Invasive Plant Management: CIPM Online Textbook

• Table of Contents
• 1.   Impacts
• 2.   Adaptive Management
• 3.   Invasion
• 4.   Inventory
• 5.   Case Study
• 6.   Risk Assessment
• 7.   Monitoring
• 8.   Management Options
• 9.   Management after Fire
• 10.  Prevention
• 11.  WMAs
• Printer friendly pdf version

Chapter 5. A Survey of Non-Indigenous Plant Species in
the Northern Range of Yellowstone National Park, 2001-2004.
An Inventory Case Study

Lisa J Rew, Bruce D Maxwell, Frank L Dougher, Richard Aspinall,
Tad Weaver and Don Despain, Montana State University

Excerpted from Final Report, available at: http://www.forestry.umt.edu/research/cesu/NEWCESU/ Assets/Individual%20Project%20Reports/NPS%20Projects/MSU/2002/ 05June%20Maxwell_Rew_Northern%20Range%20YELL_weed%20final%20report.pdf

Introduction | Methods | Results | Conclusions | Acknowledgments | References

Introduction

The United States Department of Interior National Park Service is required by law to keep the 34 million hectares designated as National Parks classified as “natural areas.” Natural areas must be “unaltered by human activities” as much as possible (U.S. National Park Service, 1996). Maintaining the Parks as “natural areas” includes removal of non-native plant species. The definition of non-native is “any animal or plant species that occurs in a given location as a result of direct, indirect, deliberate or accidental actions by humans” (U.S. National Park Service, 1996). This definition permits the user to recognize and distinguish between changes to animal and plant distributions caused by natural processes and human influences. In reality this statement needs some further clarification. “Human influence” really refers to disturbance by white settlers, more so in the past century and most specifically in the last 50 years.

Many countries have designated specific areas as “wilderness” or “natural ecosystems” and seek to preserve these in their “pristine” state, however pristine is defined. Taking this desire to “protect and retain” such areas, one can argue from the ecological purist point of view, that all non-indigenous species should be removed. However, this is currently impossible from a practical standpoint. In most cases we do not know which non-indigenous species are present within an ecosystem, their frequency or their distribution pattern; how much their distribution is changing and finally what impact they are having on the endemic ecosystem. It is only armed with all of this information that land managers can effectively target and manage non-indigenous species populations.

The language used to describe the presence and impact of non-indigenous plant species (NIS) is often very emotive: “aggressive non-indigenous plants, which spread quickly into natural areas replacing native flora and reducing habitat for native flora and fauna.”  Often the simple presence of a NIS is stated as proof enough of present or future environmental damage, particularly if it is a highly competitive species and/or if the increase in the non-indigenous species is associated with the decline of native species. However, Weaver et al., (2001) in a study of the northern Rocky Mountains found that of the 29 most commonly found exotic species the majority were intentionally introduced (e.g., Phleum pratense and Poa pratensis) and none of the most common were generally considered a noxious weed.

A number of studies have shown that when non-indigenous species are introduced to environments and ecosystems different from those in which they evolved, they may disrupt the ecosystem processes and alter biological diversity (e.g. Braithwaite & Lonsdale, 1989; Hobbs & Mooney, 1991; see Davis et al., 2000 and Mack et al., 2000 for reviews). Invasion by a new species is influenced by three factors:

1. ecosystem properties, which could be related to the level or frequency of disturbance;
2. number of propagules entering a new environment (propagule pressure); and,
3. the properties of the invading species (Lonsdale, 1999).

Davis et al. (2000) and Davis and Pelsor (2001) offer a new theory, that the fluctuation of resource availability is a key factor in controlling invasion. This theory allows for the integration of resource availability with disturbance and fluctuating environmental conditions.

Disturbance is often suggested as a key factor in enhancing the probability of NIS establishment in native plant communities. Natural disturbance has a variety of biotic and geomorphic causes including soil disturbance by fauna, weather related events such as mudflows, floods, wind, fire; and geological events such as landslides. Fire is sometimes a quasi-human disturbance if management practices suppress, contain or intentionally ignite them, or if fires are ignited accidentally or intentionally by vandals, whichever way, the natural occurrence of fires has usually been altered. Human disturbance includes construction and use of roads and trails, buildings, utility corridors and campgrounds.

As stated above, the National Park Service has a mandate to preserve the natural systems under their control (National Park Service Organic Act of 1916). There are several phases necessary to achieve this objective:

• Phase 1, creating an inventory/survey (documenting occurrence);
• Phase 2, monitoring (quantifying changes in distribution, abundance or impact); and,
• Phase 3, control or management of non-indigenous species on a large scale.

To a certain extent these phases can be performed concurrently (Fig. 5-1). The aim of the current project is Phase 1, development of an inventory/survey program.

Figure 5-1 Flow diagram for ecologically based adaptive weed management

 

The Problem with Developing an Inventory/Survey
Conducting an inventory/survey of non-indigenous plants in a large region where many of the non-indigenous species have infrequent occurrence is a difficult task. The definition of an inventory is a list of all NIS species and their locations in a delineated management area when the entire area can be observed. A survey is defined as a list of NIS species and their locations in a delineated management area when all of the area cannot be observed. A survey requires careful consideration of sampling methods. As the area of the northern range if so large and we cannot sample the entire area we are by definition completing a survey. This term will be used from now on.

Considering the ultimate use of the survey is essential in the design. In the case of the National Park Service, management of NIS is the objective, but because the NIS are relatively infrequent and spread over large areas, it will never be possible to manage all NIS or all their occurrences. Thus, a survey of the NIS and the subsequent assessment of population and metapopulation dynamics must have the objective of creating an unbiased sample in order to prioritize management of those metapopulations that pose the greatest threat to the ecosystem. An unbiased sample requires locating populations or metapopulations over the extent of the environments where they may exist. Therefore, we are reliant upon a survey that maximizes the probability of finding the non-indigenous invasive species (NIS) and simultaneously builds a data set from which models that predict NIS occurrence can be developed to ensure that we represent, through observation or prediction, all environments where the NIS may be found. It is tempting to combine survey and population assessments. If the survey is strictly a means of finding the NIS so that they can be killed, then an estimate of each metapopulation extent in the survey could serve the purpose of knowing approximately how much herbicide/hand-weeding will be required to control the observed metapopulations. However, if the intent of the survey is to maximize the potential of knowing where all of the NIS are located and subsequently using the survey to select a random sample of metapopulations to monitor for an unbiased determination of population dynamics and prioritization of management, then the survey approach that we are suggesting is most appropriate.

Methods

Study Area
Yellowstone National Park (YELL) covers an area of 899,121 ha (2,220,829 ac). Approxi­mately 1265 plant species have been recorded in YELL of which 187 (15%) are non-indigenous plant species (Whipple, 2001). The study con­centrates on the area within the northern elk winter range of the Park (152,785 ha, 377,379 ac).

Prior Knowledge of Non-Indigenous Species Occurrence in the Study Area
The relative proportional importance of the different forms of disturbance and environ­men­tal factors on non-indigenous species establish­ment and survival has not been quantified. The general perception from the National Park staff involved with NIS surveys and members of this research group was that most infestations occur close to roads, trails and human habitation. From the data collected by YELL park staff in 1998, it was calculated that 278 of 422 (66%) NIS occurrences were less than 100 m from roads or trails, and all observations were made less than 500 m from roads or trails. These data were not collected using a formal sampling strategy and the sites searched were biased by their proximity to roads and trails. Therefore, this information was treated as anecdotal and although considered, the data were not used for any subsequent analysis.

It is assumed that most of the species we are targeting are at a low frequency within the landscape and therefore collecting large numbers of observations is important to provide a reliable estimate of the species occurrence. A large sample combined with an appropriate strategy for estimating geographic distribution is also necessary if the goal is to estimate the distribu­tion of the NIS in the landscape. Survey design is, therefore, a tradeoff between collecting a sufficiently large sample to provide reliable estimates of occurrence, and using a sampling strategy that is efficient for both a) field work and b) estimating the geographic distribution of the species.

To ensure the best use of the limited funds and time available in the field, a desktop study was conducted to develop the most effective sampling regime. This was performed in ESRI ArcView ã GIS using routines developed by Aspinall and Dougher. This implemented several different sampling strategies including simple random sample, random walk, random transects, transects normal to specified linear features, stratified random sampling and regular grid sampling. Additionally, different sampling intensities were evaluated for different infestation levels (frequencies) of NIS.

The simulations and sampling strategies implemented within the GIS allowed us to evaluate which sampling strategy provides the highest number of sample points for the shortest time in the field and, also provides geographic coverage necessary for estimating distribution of the NIS. Random points or grid intersections for example, are not as efficient for collecting data as random walks or transects since time used moving from one survey location to another location is not used for data collection. Surveying along transects allows data to be collected continuously and a large sample size be generated. Additionally, surveying along transects allows changes in underlying environmental variables to be recorded. This is important for estimating the geographic distribution of the species from the sample data. This work has been published in the Biological Invasions journal (Rew et al., 2006).

If the occurrence of a target species is known to be correlated with an environmental variable, we can stratify the sampling scheme on that variable and improve our probabilities of finding the target (Hirzel and Guisan, 2002). We accepted the assumption that human disturbance in the form of roads and trails increases the chance of finding NIS, and stratified our sampling using this variable. However, to test this hypothesis we also needed to sample away from roads and trails. Therefore, transects established perpen­dicular to roads and trails were accepted as the most effective sampling methodology. The use of 2000 m transects allows the importance of other factors to be evaluated, since each transect is sufficiently long to cross a number of cover or habitat types and other environmental transitions.

Collection of Field Data
In 2001, the position of each transect was randomly selected along a road or trail, prior to arrival in the field, and ran perpendicular to roads or trails. This approach needed to be partially modified for the 2002 and 2003 field seasons to ensure a similar number of data points were collected at all distances from roads and trails. The location of transects was still randomly generated but within a set of confines:

• Starting on a road and finishing 2000 m from all roads but at all times the transect runs more than 2000 m from any known trail
• Starting on a trail and finishing 2000 m from all trails but at all times the transect runs more than 2000 m from any known road
• Starting on a road or trail and finishing 2000 m from all roads and trails.

Additionally, transect lines were generated in pairs separated no less than 100 m (when possible) and by no more than 500 m. Paired transects were used to maximize surveying time while in the field; crews would survey moving away from roads/trails on one member of a transect pair, and survey moving back towards the road/trail on the second member of the transect pair.

Non-indigenous plant species occurrence data from 2001-2003 were used to generate predic­tive models for five of the most frequently observed NIS; Phleum pratense, Bromus tectorum, Cirsium arvense, Bromus inermis, and Linaria dalmatica. In order to acquire a dataset suitable for validating the predictive models, transects for 2004 were stratified not on roads/ trails, but on the probability levels generated by these models. Transects were generated separately for areas of higher to lower probability of NIS occurrence. Apart from the different criteria in transect delineation; all survey methods used in the 2004 season were consistent with prior seasons’ methods.

Transects were walked and survey observations were made within a 10 m wide swath. I nformation was gathered when a target NIS was located, the habitat type changed or a disturbance feature was reached. The habitat classifications were based on the classifications devised by D. Despain and incorporated into the YELL GIS layers. For each NIS infestation, width and length along transect were estimated by pacing or visual determination from a central location within a patch when the patch size was small enough. When the length of the patch was too large to visually perceive or pace from a single location, the start and end of the patch length along the transect was recorded with GPS. The total length of these start and end point patches were determined by data analysis in post-processing. Patch widths were estimated up to a maximum width of 64 m.

Transect observations and location data were recorded on to GPS by two-person survey crews. Trimble Pro XR receivers and GeoExplorer3 GPS units were used and the data post-processed to improve spatial accuracy. The coordinate system and projection used was Universal Transverse Mercator (UTM) Zone 12N, WGS 1984 Datum. This projection and datum are the same as used for GIS data maintained by YELL Center for Resources, and the Greater Yellowstone Area Spatial Data Clearinghouse managed and maintained by the Geographic Information and Analysis Center (GIAC) at Montana State University .

From 2002 through 2004, all data were collected directly into a data dictionary on a Geo­Explorer3 unit that contained the same data fields as used in 2001, plus additional information on patch parameters and fields required by North American Weed Mapping Association (NAWMA, 2005). These included the location of target species, with additional information on density (in predefined classes of 0, 0-1, 1-11, 12-32, 33-100, 101-316, 317-1000 and >1000 m -2), percentage cover m ‑2, length (m) and width (m) of infestation, and spatial pattern type. Percent cover estimates were col­lected in accordance with NAWMA. Environ­mental variables included climax habitat type, dominant vegetation cover species (up to twelve species), aspect, topography and disturbance. Additional data fields included NAWMA’s “V alues at risk” and “ Ecological status of site/survey unit” and, time and date. Minor alterations in the structure of the GPS data dictionary were made with each successive field season to improve the efficiency of data collection and processing. Where these alterations resulted in discrepancies with the older data structures, the previous data files were reformatted to bring them into compliance with the newer format.

Fields that were not collected but could be added to the database at a later stage include informa­tion about the site/region, I&M network, park unit, state, county, ownership, type of survey, and non-indigenous plant species and ITIS code, all of which can be added to the database in the office.

The National Park Service has historically recorded habitat types rather than dominant vegetation/cover types. For the purpose of evaluating the environment where NIS are more likely to invade we collected data on current dominant species or successional stage, as well as the climax vegetation (habitat type).

Results

Survey Data
From 2001 through 2003, 295 transects were walked in the northern range with an overall sampled length of 529,000 m x 10 m wide (Fig. 2). Sixty-three species listed on the YELL priority list were targeted. Of these 63 species, 29 were observed in the field (Table 1). Nine of these species were observed to occur over greater than 1% of the surveyed area: Phleum pratense, Poa pratensis, Bromus tectorum, Cirsium arvense, Bromus inermis, Alyssum desertorum, Linaria dalmatica, Elymus repens and Cardaria chalepensis.

In field season 2004, 80 transects were surveyed, covering 81,900 m x 10 m (Fig. 3) Twenty-one of the 63 target species were observed in the field. Since the 2004 transects were stratified on high to lower probability of occurrence of Phleum pretense, Bromus tectorum, Cirsium arvense, Bromus inermis and Linaria dalmatica; the occurrence rates of the 21 species observed in this season are not considered to be representative of occurrence rates throughout the northern range, and are not included in the species occurrence statistics (Table 1).

NIS Distribution and Probability of Occurrence Maps
Though the survey method applied in the field was a continuous sampling along the transect, the digital representation of the survey observations was in the form of GPS points marking NIS occurrences, habitat-type transi­tions and disturbances. Using custom-made applications for Arcview (Version 3.) and Microsoft Excel, these data points were transformed into continuous linear transect data. The continuous data were then partitioned into discrete sample points at a regular 10 m sampling interval. These 10 m sample points were attributed with the presence or absence of each observed NIS, as well as associated environmental variables (e.g. burned or unburned condition, elevation, distance to nearest road, etc.). The 2001-2003 sample point data for the seven most extensively occurring NIS: Phleum pratense (Fig. 4), Poa pratensis (Fig. 5), Bromus tectorum (Fig. 6), Cirsium arvense (Fig. 7), Bromus inermis (Fig. 8), Alyssum desertorum (Fig. 9) and Linaria dalmatica (Fig. 10) have been used to characterize the distribution of NIS occurrences relative to spatial variables, such as distance to roads (Fig. 11). The analysis of the relationship between NIS occurrence and environmental variables by statistical methods, such as generalized linear regression, allows us to extrapolate those relationships throughout the northern range of YELL through predictive modeling.

Table 1. Number of observations and percentage occurrence within the area studied, for 2001-2003 data.

 

Fig. 2. Locations of all transects walked in the northern range of YELL from 2001 - 2003

 

Fig. 3. Locations of all transects walked in the northern range of YELL in 2004

 

Fig. 4. Observed presence of Phleum pratense in the northern range of YELL from 2001-2003

 

Fig. 5. Observed presence of Poa pratensis in the northern range of YELL from 2001 - 2003

 

Fig. 6. Observed presence of Bromus tectorum in the northern range of YELL from 2001 – 2003

Fig. 7. Observed presence of Cirsium arvense in the northern range of YELL from 2001 – 2003

 

Fig. 8. Observed presence of Bromus inermis in the northern range of YELL from 2001 - 2003

 

Fig. 9. Observed presence of Alyssum desertorum in the northern range of YELL from 2001 - 2003

Fig. 10. Observed presence of Linaria dalmatica in the northern range of YELL from 2001 - 2003

 

Fig. 11. Proportion of selected species observed within 100 m intervals of distance to roads, and the fitted curve for the logistic regression of NIS occurrence and distance from roads in the northern range of Yellowstone National Park for the 2001-2003 data. NIS species (top to bottom, left to right): Phleum pratense, Poa pratensis, Bromus tectorum, Cirsium arvense , Bromus inermis, Alyssum desertorum , Linaria dalmatica

 

In order to generate predictive NIS occurrence maps for the northern range of YELL, we needed to relate the presence/absence data to spatial variable data covering the entire study area. Therefore, we used environmental data derived from digital elevation maps (10 m resolution) and remotely sensed data (30 m resolution). The topographic data including aspect, elevation, slope and solar insolation (during the summer months) were calculated (the latter using the method of Swift (1976)) from the 10 m resolution digital elevation map; distance from roads and trails were calculated from data layers within the GIS database. LANDSAT Enhanced Thematic Mapper (ETM) remote sensing data, acquired July 13th 1999, were included as individual spectral bands and as an unsupervised classification layer. The unsupervised classification layer was generated using ISOCLUSTER in ERDAS Imagine and 128 classes were identified. These classes were used by Legleiter et al. (2003) to develop a land cover map of the Yellowstone watershed with accuracies of between 63 and 100% for individual land cover classes. The 128 individual ISOCLUSTER classes were used in this analysis. The 30 m resolution bands 1 to 5 and 7 of the LANDSAT ETM+ data were pan-sharpened to 15 m resolution with the panchromatic data from LANDSAT ETM+ band 8 and resampled to 10 m resolution using nearest neighbor resampling so that the resolution of the LANDSAT ETM+ data matched the resolution of the digital elevation model available for the study area. These remotely sensed data were used to provide some information on the different reflectance of the vegetation; they were used instead of the dominant vegetation GIS layer due to their finer resolution. Other environmental data, such as burned or unburned condition, and presence of trees, were obtained from the Park Service and converted into raster format as needed to work within the framework of the predictive model. The 2001-2003 data were analysed with generalized linear regression models, with binomial distribution and logit link in S-PLUS 2000. Generalized logistic models (GLM) were used because the dependent variable – the NIS species data - is binary (presence/absence) data. The best model was determined with backward stepwise procedure using Akaike’s Information Criterion (AIC), where the change in AIC value between models is used to define the “best” model, with the lowest AIC value representing the best model fit (Akaike 1977; Burnham and Anderson 1998). In our analysis we determined best model for each of the seven most frequent species using the AIC value for model selection. Probability of occurrence predictions and maps of the target species were generated using coefficient values from the GLM applied to continuous spatial variables in rasterized format, using an extension we wrote in Arcview. The extension generated the logit of the GLM by summing the product of each variable in the model and its coefficient value, plus the beta intercept value. Probability of occurrence maps were created for the seven most frequently occurring NIS: Phleum pratense (Fig. 12), Poa pratensis (Fig. 13), Bromus tectorum (Fig. 14), Cirsium arvense (Fig. 15), Bromus inermis (Fig. 16), Alyssum desertorum (Fig. 17) and Linaria dalmatica (Fig. 18). The value of each cell in the output raster, ranging from zero to one, represents the probability that the target species could be present within the area defined by that cell. In this study the raster cell size was 10 m by 10 m.

 

Fig. 12. Predicted occurrence of Phleum pratense in the northern range of YELL

 

Fig. 13. Predicted occurrence of Poa pratensis in the northern range of YELL

 

Fig. 14. Predicted occurrence of Bromus tectorum in the northern range of YELL

 

Fig. 15. Predicted occurrence of Cirsium arvense in the northern range of YELL

 

Fig. 16. Predicted occurrence of Bromus inermis in the northern range of YELL

 

Fig. 17. Predicted occurrence of Alyssum desertorum in the northern range of YELL

 

Fig. 18. Predicted occurrence of Linaria dalmatica in the northern range of YELL

Conclusions

The aim of this project was to determine the best sampling approach for a NIS inventory/survey in YELL, and complete the field survey. We have achieved this, and a bit more. The best sampling method was evaluated using a computer simu­lation model and some field sampling in 2001. The most efficient and accurate method to represent a species’ frequency over the landscape as a whole was used. This method entailed randomly-stratified transects on roads and trails, i.e. transects started on roads or trails and finished 2000 m from either or both of them. By using this unbiased sampling approach it was possible to develop probability of occurrence models for individual species which provide information on the entire area of interest. We believe such probability of occurrence maps will be useful to managers, we hope we are right.

The probability of occurrence maps provide information on areas which should be searched first when looking for new populations; and, they pro­vide data on populations which could be used for the next phase of NIS management, monitoring. As with the inventory/survey phase it is not possible to monitor all populations. We do not believe that all populations of a par­ticular NIS are invasive, increasing in spatial extent and density, and having a negative impact, in all the environments in which they occur. Therefore, we need a better under­standing of where particular species’ popula­tions are increasing/having and impact and where they are not having such a negative effect. Thus, by sampling a number of popula­tions from different environments we can get a better and more informative understanding of this issue. This should be the next phase of this project.

The current project has received a great deal of interest in the scientific and management arena. Over the four-year period we have set up simi­lar sampling systems in other areas including Bighorn National Recreation Area, Gallatin National Forest and Kootenai National Forest. Two scientific manuscripts have been published, and the full text of this report is available online, as detailed below.

Acknowledgements

This project was funded by the Yellowstone Inventory and Monitoring program, under the supervision of Cathie Jean. Sincere thanks to Lane Cameron for initiating this project and his continued interest and support, even after leaving for the Channel Islands National Park. Thanks also to Mary Hektner and Roy Rankin for their non-indigenous plant species and logistics insight; to Jennifer Whipple for her help with plant identification; and Ann Rodman and the GIS lab in Yellowstone National Park for sharing their GIS data layers.

Special praise and thanks is reserved for the field crew, particularly Jerad Corbin, Mara Johnson and Amanda Morrison who worked on the project for a couple of summers, but also to Judit Barroso, Matthew Hulbert, Jeremy Gay, Rebecca Kennedy, Erik Lehnoff, Ben Levy, Nathaniel Ohler, Charles Repath, John Rose and Kit Sawyer. All of them exerted considerable energy and enthusiasm for plant identification, data collection and hiking all sorts of terrain in all sorts of conditions; the project would not have achieved as much without such dedication.

References