0665-B1

Classifying U.S. Forest Inventory and Analysis Data to the National Vegetation Classification Standard

Shannon Menard^[1], Don Faber-Langendoen, John Vissage, Pat Miles and Brad Smith

Abstract

NatureServe and the U.S. Forest Service Forest Inventory and Analysis program (USFS FIA) are collaborating on methods to link FIA forest plot data to the U.S. National Vegetation Classification (USNVC) standard at the alliance level. An alliance is defined as a physiognomically uniform group of plant associations sharing one or more dominant or diagnostic species usually found in the upper-most or dominant stratum. Although similar in scale to the Society of America Foresters' forest cover types (the current FIA standard), alliances better reflect floristic and ecological conditions. A pilot study was initiated in Province 212 in MN and WI using 1 389 forested, stocked FIA "plots" (where "plot" is the aggregated set of subplots that share the same FIA condition) collected from 1998-2001. Thirty-seven forest alliances occur in the region. Relative Importance Value based on density and basal area was calculated by tree species for each FIA plot. The datasets were divided based on natural vs. planted and mesic/xeric vs. hydric classes. Within each class, Forest Groups were recognized based on a combination of "Indicator Species". Alliances were nested within these groups, and keys were developed that first assigned plots to groups, then to alliances. Non-metric multidimensional scaling and flexible beta cluster analyses were used to begin testing the proposed assignments of plots to groups and alliances. An algorithm to assign plots to alliances is under development. This pilot project lays the groundwork for linking FIA plot data to the USNVC across the country, thus improving FIA ability to monitor status and trends of the nation's forests.

Introduction

The U.S. Forest Service (USFS) maintains information concerning the nation's forests through its Forest Inventory and Analysis (FIA) program. FIA has been operating since the 1930s and represents the most comprehensive effort to collect and analyze forest data in order to monitor the status and trends of forests in the United States (Smith 2002). For forest classification purposes, FIA has traditionally used Society of America Forester's (SAF) cover types, which depend primarily on dominant tree species (Eyre 1980). A computer algorithm links FIA data to these cover types, providing a consistent classification of FIA data (Arner et al., in prep.).

FIA data were historically used primarily to define current and potential timber resources; however, during the last 2 decades there has been a shift in the interpretation of the mandate for FIA to a more holistic view of monitoring ecosystems (Smith 2002). This demand necessitates the use of a more ecologically and floristically based classification system in addition to SAF forest types. Here, the potential value of using the U.S. National Vegetation Classification System (USNVC, Grossman et al. 1998) to meet this need is assessed.

The USNVC is maintained by scientists from NatureServe and the Natural Heritage Network, working in partnership with the Ecological Society of America, federal agencies, and others (Jennings et al. 2002). This classification is the national standard for U.S. federal agencies (FGDC 1997). It is a hi erarchically-based classification with the upper levels defined primarily by physiognomy and the lower levels by floristic similarities. The alliance floristic level is defined as a physiognomically uniform group of plant associations sharing one or more dominant or diagnostic species, which as a rule are found in the upper-most stratum of the vegetation (Grossman et al. 1998). Alliances are similar in scale to SAF forest types and represent the best USNVC unit to link to FIA data, which only include woody data.

A pilot study was initiated to examine the utilities and methodologies of linking FIA data to USNVC alliances within the Minnesota and Wisconsin portion of the Laurentian Mixed Forest Province (Province 212; Bailey et al. 1994). Specifically the objectives of this study are to (1) develop a method to link FIA plot data to alliances through the use of keys and algorithms similar to the stocking algorithm already being used by FIA, and (2) identify issues regarding how best to link FIA data to alliances and suggest alternative methods for collecting FIA data and describing alliances that would improve this linkage for a nation-wide algorithm.

Methods

Data preparation

The plot data used for this project were derived from the FIA Phase 2 (P2) annular plots, which only include data from mature trees to seedlings. FIA plot data are structured in a complex manner, based on whether the 4 subplots within a plot share the same "condition" class (USDA-FS 2000). Condition reflects basic differences in stand condition, such as forest versus non-forest and water versus terrestrial. For this reason, the subplots were aggregated based on condition class into "plot-condition" (as FIA does for all of its reporting). Thus, although there were 760 original plots for this pilot study, many had multiple conditions, resulting in 1389 "plot-conditions" (hereafter referred to simply as plots). This large plot dataset was produced after filtering the plots through a decision matrix to determine which ones qualified as forest plots (as only forest alliances are considered in this pilot). All plots were initially filtered according to whether or not they were forestland and stocked; if these conditions were not met, those plots were eliminated (Figure 1).

Figure 1: Decision matrix from pilot study to determine plot placement in the primary datasets.

A total of 37 USNVC forest alliances are known from Province 212 in MN and WI (Drake and Faber-Langendoen 1997, NatureServe 2002). Nine additional woodland alliances are also known in this region, however, they are relatively rare, and thus no additional criteria were considered to help distinguish them in this study. Because alliances also typically reflect site and disturbance factors and not just simple canopy dominance, two criteria were used to subset the data in order to improve the correlation between plots and alliances. First, the data were divided into artificial (planted) and natural plots. Second, each of these datasets was divided by two broad physiographic classes attributed to plots by FIA (hydric and mesic/xeric). This created four potential primary datasets for further analyses (Figure 1). The majority of the 1389 plots (78.4%) were classified as natural/mesic-xeric stands, 5% as artificial/mesic-xeric and 16.6% were designated natural/hydric stands. There were no artificial/hydric plots. Once these datasets were identified, relative importance value (RIV) was calculated for each species by strata for all plots in each dataset. RIV is the sum of relative density and relative basal area, divided by 2 with tree and sapling data combined into one RIV per species per plot.

Keys to Forest Groups and Alliances

Preliminary data analyses of the above three datasets indicated that the further subdivision of the datasets would be helpful in assigning plots to alliances. Each of the datasets was subdivided into Forest Groups based on a shared set of "Indicator Species" derived initially from qualitative information. The resultant Forest Groups are similar in concept to the SAF Forest Type (Eyre 1980). The cumulative RIV of each species within given Indicator Species groups was calculated for each plot, and the proportion of these Indicator Species was used in a key to assign the plot to a Forest Group. Each Forest Group contains a unique subset of alliances. Thus, further criteria based on the RIVs of indicator and dominant species were used in a key to assign each plot to an alliance. Criteria for the keys were taken from information provided in the alliance descriptions (Drake and Faber-Langendoen 1997, NatureServe 2002). These expert-driven keys were then compared to results from quantitative analyses.

Data Analyses

Non-metric multidimensional scaling (NMS) and detrended correspondence analysis (DCA) were explored using PCORD (McCune and Medford 1999). Species on <2% of the plots and with <40% RIV were removed from the 3 datasets to remove the impact of those relatively uncommon species on the analyses. NMS showed less visual "skewness" in the large datasets and was used to ordinate the datasets. Flexible Beta clustering (McCune and Medford 1999) was used to assess similarities among the plots. After identifying clusters of similar plots, their unifying vegetation and environmental characteristics were summarized. These characteristics then were used in conjunction with the qualitative approach above to assess the definitions of alliances and the criteria used in the keys.

Results

Ten Indicator Species groups were defined based on the qualitative analyses (Table 1). From these, 12 Forest Groups with unique sets of alliances were recognized (3 wetland, 3 plantation or semi-natural, and 6 natural uplands; Table 2). Over 99% of the plots could be assigned to a Forest Group based on the keys created to classify plots into Forest Groups. Within the various Forest Groups, separate keys to each of the 37 alliances were then created based on the relative percentages of indicator and dominant species and their RIV thresholds (Box 1). Over 99% of the natural, mesic-xeric plots and natural hydric plots were linked to an alliance. All of the plantation and semi-natural plots were likewise assigned to an alliance, although several new alliances appear to be needed to cover the variability within these Forest Groups.

Table 1: Example of five groups of Indicator Species used to define Forest Groups. The species numbers correspond to those used by USFS FIA project (Miles et al. 2001). Additional groups (not shown) include "Plantation Conifers (introduced)," "Introduced Hardwoods," "Floodplain/Swamp Hardwoods," "Northern Oaks," and "Other Hardwoods."

Table 2: Example of Forest Groups within Bailey Province 212 and the USNVC Alliances nested within them.

Ordination and cluster techniques show that plots tended to separate according to Forest Groups. Results from NMS Ordination demonstrate the separation of plots into two types of Forest Groups, those that have a more boreal and/or northern conifer component compared to those with more temperate hardwoods (Figure 2). Likewise, the cluster dendrogram shows the plots clustering according to individual Forest Groups. Approximately 40 clusters were identified for the natural, mesic-xeric group of plots, ranging in size from 7 to over 150 plots. Within each of these clusters approximately 54-100% of the plots fell within the same Forest Group (Table 2, Figure 3). Those in a cluster not designated the same Forest Group were often designated to very closely related alliances using the alliance keys within a given Forest Group. For example, in cluster V, two of the stands classified to the Northern Hardwoods Forest Group were assigned to an oak dominated alliance using that key. Based on these clusters, the plots were reclassified to the Northern Oak Forest Group, where that oak alliance is listed. These results suggest that further analyses will help to refine the keys to Forest Groups and alliances, and ultimately provide clear quantitative descriptions of alliances.

Figure 2: NMS results showing the distribution of Forest Groups typically more boreal in nature versus those more temperate. The line shows the general groupings of natural, mesic-xeric plots into these categories.

Figure 3: Excerpt from the dendrogram results using Flexible Beta clustering technique for natural, mesic-xeric plots. Five of approximately 40 total clusters are displayed with Forest Group assignments by plot.

Discussion

This pilot study demonstrates that USNVC alliances can be successfully linked to FIA P2 annular plots using the tree data available in the FIA database. There were several challenges to overcome in linking FIA plots to alliances. The sheer size of the datasets can make the analyses somewhat unwieldy and difficult to interpret. The division of the plots into Forest Groups based on Indicator Species and the use of keys to alliances based on qualitative information from alliance descriptions helped provide a first approximation for assigning FIA plots to alliances. Subsequent quantitative analysis using ordination and cluster analysis (Figures 2, 3) provided an important test of these keys.

The use of a Forest Group level above the alliance is comparable to the SAF "Forest Type Group" above the Forest Type (Eyre 1980), which is also used by FIA to summarize inventory data. The Forest Group level, if developed further within the USNVC structure, could provide FIA and other federal agencies a broad scale floristic unit suitable for regional and national analyses. The alliance keys under each Forest Group will be used to develop an algorithm similar to the stocking algorithm already being used by FIA (Arner et al. in prep.), and which will allow FIA to link their plots to alliances directly. Both the alliance and the Forest Group level would provide FIA with the needed tools to report on the nation's forests.

There are some issues that should be addressed with respect to assigning FIA plot data to USNVC alliances. Alliances in this region have been primarily defined based on qualitative information (Drake and Faber-Langendoen 1997). Thus, it is often not clear how much variability in the RIV for a set of dominant or diagnostic tree species is allowed within the confines of the alliance description. Alliance definitions will need to be revisited to determine how best to draw distinctions among similar types using quantitative information.

Another issue is that of distinguishing wetland versus upland alliances. FIA only considers plots that are saturated throughout most of the year (e.g. swamps) to be hydric, whereas plots found in the floodplain or bottomlands are considered mesic (USDA-FS 2000). By contrast, the USNVC separates all wetland types from upland types, and then further distinguishes wetland alliances based on the hydrologic regime of Cowardin et al. (1979, in Grossman et al. 1998). Because the FIA mesic/xeric class covers both wetlands and uplands, wetland alliances may occur in both this dataset and in the hydric dataset, and thus complicate the assignment of plots to alliances. FIA does allow for a finer classification of their plots beyond xeric, mesic, and hydric into categories such as swamp, floodplain, etc (USDA-FS 2000); however, this information was only available for approximately 64% of the plots. More consistent use of the finer physiographic classes by FIA would improve the link between plots and wetland alliances.

Conclusions

FIA P2 plots can be linked to the USNVC standard classification at the alliance level. Keys to the alliances were developed from information in the alliance description and quantitative plot information. The use of a Forest Group level above the alliance helped to partition the large FIA datasets into smaller, more manageable subsets, both for the purposes of writing these keys and for further quantitative analyses. Quantitative criteria used in the keys can inform the computer algorithm that would automate the assignments; however, further analyses and field checks are still needed to completely validate the keys.

A few changes in some of the FIA plot collection methods, such as consistent use of the finer physiographic class modifiers and quality checks of species information to identify possible misidentifications will help refine the link between FIA P2 data and the USNVC. Likewise, some additions to alliance descriptions to reflect the quantitative information provided by FIA will help other agencies and users of FIA data to repeat these analyses using alliances information.

Preliminary tests of the approach using quantitative analyses such as NMS and Flexible Beta suggest that the criteria used in the key can be improved through such analyses. Furthermore, these analyses represent an opportunity to test and refine the alliance concepts themselves. This pilot study provides a strong basis for further work linking USNVC alliances to FIA P2 plots throughout the lower 48 United States, thus enabling FIA data to be easily shared with other federal agencies already using the USNVC to map and delineate vegetation communities.

Literature Cited

Bailey, R.B., P.E. Avers, T. King, and W.H. McNab (eds). 1994. Ecoregions and subregions of the United States. U.S. Forest Service, Washington, DC. (scale 1:7,500,000, with supplementary table of map unit descriptions).

Drake, J. and D. Faber-Langendoen. 1997. An alliance level classification of the vegetation of the Midwestern United States. The Nature Conservancy, Midwest Conservation Science Department, Minneapolis, Minnesota, USA.

Eyre, F. H., editor. 1980. Forest cover types of the United States and Canada. Society of American Foresters, Washington D.C., USA.

Federal Geographic Data Committee. 1997. Vegetation classification standard, FGDC-STD-005. Web address: http://www.fgdc.gov/standards/status/sub21.html

Grossman, D.H., D. Faber-Langendoen, A.W. Weakley, M. Anderson, P. Bourgeron, R. Crawford, K. Goodin, S. Landaal, K. Metzler, K.D. Patterson, M. Pyne, M. Reid, and L. Sneddon. 1998. International Classification of Ecological Communities: Terrestrial Vegetation of the United States. Volume I: The National Vegetation Classification System: Development, Status, and Applications. The Nature Conservancy, Arlington, VA.

Jennings, M., O. Loucks, D. Glenn-Lewin, R. Peet, D. Faber-Langendoen, D. Grossman, A. Damman, M. Barbour, R. Pfister, M. Walker, S. Talbot, J. Walker, G. Hartshorn, G. Waggoner, M. Abrams, A. Hill, D. Roberts, and D. Tart. 2002. Standards for the associations and alliances of the U.S. National Vegetation Classification. The ESA Vegetation Classification Panel, Version 1.0 May 2002.

McCune, B., and M.J. Mefford. 1999. PC-ORD. Multivariate Analysis of Ecological Data, Version 4. MjM Software Design, Gleneden Beach, Oregon, USA

Miles, P. D., G. J. Brand, C. L. Alerich, L. F. Bednar, S. W. Woudenberg, J. F. Glover, and E. N. Ezzell. 2001. The forest inventory and analysis database: users manual version 1.0. USDA Forest Service General Technical Report NC-218.

NatureServe. 2002. International Classification of Ecological Communities: Terrestrial Vegetation. Natural Heritage Central Databases. NatureServe, Arlington, VA.

Smith, W. B. 2002. Forest inventory and analysis: a national inventory and monitoring program. Environmental Pollution 116: S233-S242.

U. S. Department of Agriculture, Forest Service. 2000. Forest inventory and analysis North Central national core field guide: field data collection procedures for phase 2 plots, version 1.4. USDA-FS, North Central Research Station. Internal report. On file with: USDA-FS, North Central Research Station, Forest Inventory and Analysis, St. Paul, MN.