Historic, archived document Do not assume content reflects current scientific knowledge, policies, or practices. Sui! ; A99.9 F764Un Racku Great Plains outhwest aoe wy Research Note RM-528 USDA Forest Service May 1994 Rocky Mountain Forest and Range Experiment Station Robustness of the Point-Line Method for Monitoring Basal Cover John E. Mitchell’, Ward W. Brady’, and Charles D. Bonham? The point-line method is a robust estimator of basal cover with respect to population distribution, average plant size, and the non- randomness of a disturbance. A graphical modeling approach can evaluate the overall effectiveness of any vegetation inventory proce- dure. Keywords:range condition, range inventory, robust measures Introduction Range condition classification historically has been used to delineate rangeland health (Joyce 1993). Over the past decade, the Department of Defense (DOD) has initiated a program, entitled Land Condition-Trend Analysis (LCTA), to moni- tor the response of soils and vegetation to training activities on military lands (Tazik et al. 1992). The LCTA method closely follows techniques for esti- mating range condition by providing erosion indi- cators and measures of bare soil and plant cover by species. These data are obtained from field plots containing a 100-m point-line transect along which a point reading is made every meter. The point-line method has been described by Bonham (1989). "Range Scientistwith the Rocky Mountain Forestand Range Experiment Station. Headquartersisin FortCollins incooperation with Colorado State University. *Professor, Schoolof Agribusinessand Environmental Resources, Arizona State University, Tempe, AZ85287. °Professor, Departmentof Range Science, Colorado State University, Fort Collins, CO80523. Disturbance patterns due to military training activities are mostly non-random. Vegetation per- turbation from bivouac activities probably are non- random in relation to LCTA point-line transects because of the size of platoon and company biv- ouac areas. For example, the trampling pattern within and around a medium-sized army tent is approximately 10 m by 20 m (personal observa- tion). Tread marks left by tracked vehicles during field exercises are a dominant form of non-random disturbance. Monitoring designs should be stable, powerful, and robust if they are to detect meaningful changes in vegetation with an acceptable error (Green 1979). Stable designs have an acceptable probability of false conclusions concerning vegetation change (Type-I error); powerful designs have an acceptable likelihood of actually detecting vegetation change (complement to Type-II error). Robust methods are those that produce data that are not influenced by extraneous factors (i.e., factors not considered when collecting data), particularly when estimators are based upon small samples. Frequency measures are not robust, for example, because they are af- fected by plot size and shape (Bonham 1989) and plant size (Cook et al. In press). The objective of this study was to evaluate the robustness of the LCTA point-line method using a graphical modeling approach. Specifically, we wanted to determine the method’s ability to mea- sure changes in basal cover in relation to aggregated disturbance patterns and non-random vegetation distributions. Methods Evaluating the robustness of the point-line tech- nique for monitoring changes in basal cover is a formidable task in the field; the cost of creating treatments is prohibitive and numerous replica- tions are needed to ensure adequate precision. Therefore, a computer simulation approach was used because it allowed easy measurements of specified population distributions with a large number of replications. We used Turbo Pascal (Version 6.0)* as a pro- gramming language to develop graphic computer models of a simulated shortgrass steppe commu- nity on the Pinon Canyon Maneuver Site (PCMS) in southern Colorado (Shaw et al. 1989). The PCMS serves as a training area for mechanized infantry and armor forces located at Fort Carson, Colorado. The shortgrass community was defined in terms of the dominant species, blue grama (Bouteloua gracilis (H.B.K.)Lag.), with an initial basal cover of 12%. Our rationale was that a monitoring system should be able to follow the dynamics of blue grama, which typically constitutes 50-90% of the community biomass (Shaw et al. 1989), if it is to be minimally effective. We used a scale of 1 cm per pixel to generate vegetation models. Such a scale portrayed a com- munity of 10.24 m by 7.68 m on one screen of a 8514A monitor. Thus, a simulated community large enough to contain a 100-m transect, randomly situated, was represented by 11 separate, but linked, graphic screens. 4Theuse oftradeandcompanynamesis forthe benefitofthe reader, such use doesnotconstitute an officialendorsementorapprovalofany service orproductby the U.S. DepartmentofAgnculture to the exclusion ofothers thatmay be suitable. Each of the 100 points along a point-line transect was expressed as a single pixel. Although the resultant “points” were 1 cm’, they still provided an unbiased estimator of cover because an observa- tion was always dichotomous; i.e., each pixel was always completely covered by a plant or not cov- ered at all. Therefore, the point was no less dimen- sionless than the vegetation. Treatment effects examined how average plant size, plant distribution, and disturbance mode in- fluenced the ability of the point-line method to predict basal cover. Two plant sizes, 8 and 12 cm in diameter with standard deviations of 2 and 3 cm, respectively, were simulated. These size distributions were based upon field measurements of shortgrass steppe veg- etation from comparable locations because specific plant size data were unavailable from PCMS.° All plants were circular. Three plant distributions were generated: Ran- dom, moderately contagious, and highly conta- gious. We created moderately and highly conta- gious distributions by defining random clusters (mean radius of 1 m,s = 0.5 m) within the simulated sampling space. Plants were given a 50% probabil- ity of being inside the cluster circles in the moder- ately clumped distributions; this probability in- creased to 95% for the high degree of clumping. The circles and distributions were devised follow- ing the algorithm shown in figure 1. We used three modes of disturbance to create decreases in basal cover of the simulated popula- tions. Disturbance was random when each plant had an equal chance of being eliminated. The second and third disturbance modes assumed that plants were being killed by the actions of a tracked vehicle, the M-1 main battle tank. Loaded M-1 tanks weigh about 55,000 kg and can cause substan- tial damage to grassland vegetation (Shaw and Diersing 1990). Each M-1 tank tread is 0.63 m wide, and the distance between the treads is 2.24 m. Tank “tread marks” were randomly overlaid on the sam- pling space until they “destroyed” sufficient plants to bring blue grama basal cover down to the desired treatment level. The probability that a plant was eliminated when passed over by a tank tread was 0.5 for the second disturbance mode and 1.0 for the third disturbance mode. 5 Unpublished data from SchoolofAgnibusiness and Environmental Re- sources, Arizona Siate University, Tempe. Onfile with second author. Locate P,,,atL,,; within perimeter Locate plant n (P,,) at random location (L,) Yes Is community basal cover = 12%? No Perimeter of cluster determined Hrad = 1.0 Orad =0.5 Pick random number, Osrsi Yes No Locate F,,, at new L441 outside cluster Define new cluster at P,,., with Hrad = 1-0 Gpaq = 9.5 No Is basal cover 12%? Yes * P(c) = probability of é random plant falling in cluster: P(c) = 0.5 or 0.95 C stor ) for the two non random distributions. Figure 1.—Flow diagram of algorithm for locating plants in a simulated shortgrass steppe following a contagious distribution. The algorithm for situating tread marks con- formed to the following sequence: (1) A “tank” would enter the sampling space at a random point on its perimeter and in a random direction; (2) when the tread marks encountered a plant, the algorithm would ascertain whether the plant was eliminated from the population (P = 0.5, 1.0); (3) the tank would then move on in a straight line until it came upon another plant, exited the sampling space, or reached the desired reduction in basal cover; and (4) once a tank exited the sampling space, the process repeated itself until the reduced plant cover was attained. Simulations followed the Monte Carlo method of approximating the solution to a problem by sampling repetitively from a random process (Springer et al. 1968). Each plant size, vegetation pattern, and disturbance mode combination of simu- lated shortgrass vegetation was sampled with 10,000 randomly located “permanent” LCTA line transects. Basal cover was estimated for each transect from the 100 evenly spaced points. Resampling and disturbance took place iteratively until reduced population basal covers of 10%, 8%, 6%, 4%, and 2% were reached. We evaluated the robustness of the simulated point-line method by comparing its power to detect different cover changes across a range of sample sizes. Sample size was adjusted by combining from 1 to 10 transects. Grouping of transects assumed homogeneity within the sample space. Power, the probability of detecting an actual change in basal cover (1.0 - P [Type-II error]), was estimated from comparisons of areas under actual distribution % Cover Figure 2.—Type-l and Type-ll errors (2-tailed) portrayed for a hypothetical population where plant basal cover has decreased from 12% (time 1) to 8% (time 2). The area under the normal curve (1 - 8) corre- sponds tothe sampling design’s power. curves resulting from the sampling-disturbance- resampling iterations (fig. 2). A more detailed dis- cussion of power curves has been provided by Tanke and Bonham (1985). The effects of plant size and disturbance mode on robustness of the point-line method was studied by observing a decrease in basal cover from 12% to 10% and 6%, respectively. As a comparison, the effect of population non-randomness was exam- ined by observing basal cover decreases from 12% to 10% and 8%. If the null hypothesis of no treat- ment effects was to be accepted, the common re- duction to 10% basal cover would result in similar power curves; moreover, the reduction to 8% cover would be intermediate to the two sets of curves testing reductions to 10% and 6%. Type-I errors were set at approximately P = 0.05 for all power calculations. Error could not be set pre- cisely at 0.05 because discrete populations were used. Results and Discussion The point-line method appeared to be robust with respect to both mean plant size and mode of disturbance throughout the treatment ranges. The probability of detecting a 2% decrease in basal cover of shortgrass steppe, from 12% to 10%, was 0.06

12cm-Ran 4, 801-4) < 12cm-Mod J W Ad os { ~~ 12cm-Sev | ac { 5 40% Ss } a No. Point-Line Transects Figure 3.—Effect of plant size and disturbance pattern on power curves for a decrease in basal cover from 12% to 10% (solid lines) and 6% (dashed lines) of simulated shortgrass steppe. a =0.05 (the exact error could not be specified for a discrete popu- lation). 100 | —__ > S poe ee 80 Ne i, ye ° ea = Ue x Ze om 507- BLOT £ poe 3 es io Wee S 407 F é eee | a; = Random | | + Mod | | High 1 2 3 4 5 6 7 8 9 10 No. Point-Line Transects Figure 4.—Effect of three plant distributions on power curves for a decrease in basal cover from 12% to 10% (solid lines) and 8% (dashed lines) of simulated shortgrass steppe.a =0.05 (the exact error could not be specified for a discrete population). large number of iterations in the Monte Carlo method; thus, this range in prediction represents a random error in the power rating predictive model (Gelb 1974). The decrease in random error at the extremes is expected, given that the power of the test follows a logistic function (Tanke and Bonham 1985). As a result, all power curves closely parallel each other as the possibility of detecting a population change converges on 0% and 100%. Conclusions Power analysis can demonstrate to resource managers how well a monitoring system can detect a pre-established change in ecosystem state. For a monitoring design to be useful, however, it must be robust across an unknown number of ecosystem variables. The point-line method, regardless of other shortcomings, is a robust estimator of basal cover with respect to population distribution, average plant size, and the non-randomness of a disturbance. Outside the limits of this particular sampling method, we conclude that a graphical modeling approach has excellent utility for evaluating the overall effectiveness of any vegetation inventory procedure. It allows an investigator the opportu- nity to set plant community parameters, compare sampling methods, and determine sample ad- equacy. Literature Cited Bonham, Charles D. 1989. Measurements for ter- restrial vegetation. New York, NY: John Wiley & Sons. 338 p. Cook, John W.; Brady, Ward W.; Aldon, Earl F. The effect of grass plant size on basal frequency estimates. Grass and Forage Science [in press]. Fisser, H.G.; Van Dyne, G.M. 1966. Influence of number and spacing of points on accuracy and precision of basal cover estimates. Journal of Range Management 19:205-211. Gelb, Arthur, ed. 1974. Applied optimal estima- tion. Cambridge, MA: The MIT Press. 374 p. Green, Roger H. 1979. Sampling design and statis- tical methods for environmental biologists. New York, NY: John Wiley & Sons. 257 p. Joyce, Linda A. 1993. The life cycle of the range condition concept. Journal of Range Manage- ment 46:132-138. Shaw, R.B.; Anderson, S.L.; Schulz, K.A.; Diersing, V.E. 1989. Plant communities, ecological check- list, and species list for the U.S. Army Pifion Canyon Maneuver Site, Colorado. Science Series No. 37. Fort Collins, CO: Department of Range Science, Colorado State University. 71 p. Shaw, R.B.; Diersing, V.E. 1990. Tracked vehicle impacts on vegetation at the Pinon Canyon Ma- neuver Site, Colorado. Journal of Environmental Quality 19:234-243. Springer, Clifford H.; Herlihy, Robert E.; Mall, Robert T.; Beggs, Robert I. 1968. Probabilistic models. Vol. 4: Mathematics for management series. Homewood, IL: Richard D. Irwin, Inc. 301 p. Tanke, William C.; Bonham, Charles D. 1985. Use of power curves to monitor range trend. Journal of Range Management 38:428-431. Tazik, David J.; and six others. 1992. U.S. Army Land Condition-Trend Analysis (LCTA) plot in- ventory field methods. USACERL Technical Report N-92/03. Champaign, IL: U.S. Army Corps of Engineers, Construction Engineering Research Laboratory. 62 p. The United States Departmentof Agriculture (USDA) Forest Service is a diverse organization committted to equal opportunity in employ- mentand program delivery. USDA prohibits discrimination on the basis of race, color, national origin, sex, religion, age, disability, political affiliation, and familial status. Persons believing they have been discrimi- nated against should contact the Secretary, US Departmentof Agricul- ture, Washington, DC 20250, orcall 202-720-7327 (voice), or 202-720- 1127 (TDD).