Previous PageTable Of ContentsNext Page

2 LCTA Data Analysis

There are three primary objectives for the analysis of LCTA data:

The standard LCTA design at an installation consists of approximately 200 permanent 100-m point intercept transects randomly located across an installation. Data on ground and aerial cover are collected every 1 to 5 years (Diersing, Shaw, and Tazik 1992). The relative cover of each species by plot was used in the multivariate analysis to separate and describe the plant communities.

The multivariate analyses of the vegetation was conducted for each installation and for all five installations combined. The results presented below are from the combined analyses.

Multivariate Statistical Procedures

The multivariate classification procedure follows that described by McLendon and Dahl (1983). Principal component analysis (PCA) was used to establish an unbiased initial grouping of the transects. A stepwise discriminant analysis (SDA) was conducted for each installation separately to reduce the initial number of groups. These analyses reduced the number of potential groups from 48 per installation to 15 to 20 statistically different groups per installation (Figure 1). The resulting 88 groups and 922 transects were then entered into a combined SDA, which resulted in a final classification of 68 groups (plant communities) and placement of each of the 922 transects into its respective group.


Figure 1. Flowchart of the multivariate statistical analysis of LCTA
vegetation data from five installations.

Results of Multivariate Analysis

Analysis of the 1989 LCTA vegetation data sets combined for the five installations resulted in the separation of 68 statistically distinct communities, including 14 woodland, 9 shrubland, 38 grassland, and 7 early-seral communities. On an intra-installation basis, Fort Bliss had the most heterogeneous and Fort Hood the most homogeneous vegetation (Table 1). The earlier installation-only SDA had identified 20 communities at Fort Bliss (Figure 1). However, when placed in the combined SDA, five of these communities were not statistically different from other communities at Fort Bliss, therefore the 20 groups were reduced to 15 in the combined SDA (Table 1). At the same time, the vegetation along some transects at Fort Bliss was more similar to that in Fort Carson (5) and Fort Hood (1) communities than any of the 15 remaining Fort Bliss communities (Table 1).

Stepwise discriminant analysis also provides a measure of the statistical difference among the groups. By averaging these difference values, weighted by number of transects per group, an average difference value can be calculated that compares differences within and between installations (Table 2). Although Fort Hood had the most homogenous vegetation based on number of communities, Yakima TC had the most homogenous vegetation based on statistical differences among its communities (diagonal values, Table 2). Fort
Table 1. Summary of the results of the multivariate analysis of 1989 LCTA vegetation data combined for all five installations.

Riley had the statistically more diverse vegetation (average F-ratio of 150, Table 2), perhaps reflecting the strong ecological differences between the woodlands and grasslands there.

Differences in vegetation among installations overall was greatest for Yakima TC (highest off-diagonal values, Table 2). Yakima is located farthest away from the other four installations and its vegetation is dominated by cold desert species. The vegetation of Forts Bliss, Carson, and Hood are approximately equal in similarity. These three installations are in semiarid regions and share a number of the same species.

This multivariate analysis provides an excellent method for comparing vegetation across broad geographical and ecological scales. But it also allows for detailed comparisons at the landscape scale within an installation. Table 3 represents relative cover of the major species of 14 plant communities, 5 woodland and 9 grassland, at Fort Riley, KS.

Table 2. Average F-ratio values testing significance of statistical differences in vegetation within (diagonal) and between (off-diagonal) installations.*

*Results are taken from the stepwise discriminant analysis of the 1989 LCTA data set combined for all five installations.
Table 3. Relative cover values (%) of the major species within each of the 14 Fort Riley plant communities, as identified by multivariate analysis of the 1989 LCTA data.

*All codes are from the U.S. Department of Agriculture National List of Scientific Plant Names, Volume 1, List of Plant Names.

Discriminant analysis can also be used to evaluate temporal changes in vegetation. The LCTA data sets are especially useful in this task. To conduct this analysis, we entered data from the same transects but from different years. An analysis for Fort Riley was conducted using 1989 and 1994 data. The Ange/Sonu-Bocu community (G1) in 1994 was statistically different from what it was in 1989. This was largely the result of an increase in tall dropseed (Spas) and decreases in big bluestem (Ange) and sideoats grama (Bocu) from 1989 to 1994 (Table 4). At the same time, there was a shift in the Ange/Sonu-Spas (G2) community toward the Ange/Sonu-Bocu (G1) community. This was the result of decreases in big bluestem (Ange), Indiangrass (Sonu), and sideoats grama (Bocu) during the 5 years (Table 4).

Table 4. Changes in species composition (% relative cover) in the grassland communities at Fort Riley 1989 and 1994.

Plans for Statistical Development

We plan to continue the analyses of LCTA data sets in four primary task areas. First, we plan to conduct more detailed analyses of spatial patterns at each of the five installations. In particular, we are interested in analyzing for patchiness within individual transects, and using this information in the development of the spatial module of our simulation model. Second, we plan to conduct further analyses of the temporal aspects of the vegetation comparing changes over time across the broad geographical and ecological scales encompassed by these LCTA data. We would also like to continue the temporal analyses at each installation as LCTA data become available for more years. Third, we would like to test the accuracy of existing vegetation maps at each of the installations by use of LCTA data. If the accuracy of the current vegetation maps is found to be unacceptable, we would like to develop new maps based, in part, on our analyses of LCTA data. Fourth, we would like to expand the analysis of LCTA data to other installations.

Previous PageTable Of ContentsNext Page