This file was created by scanning the printed publication. Errors identified by the software have been corrected; however, some errors may remain. Complexity of Sampling Multiple Resources 1 W.E. Frayer2 Abstract-Experience with many inventories for single and multiple resources has resulted in suggestions to be considered when sampling multiple resources. Discussion begins with the design of the sampling unit and continues for such items as sampling design, sampling over time and projections of future status. Specific recommendations are given for each area discussed. Many inventories are originally designed to monitor status and trends of a single resource. There have been inventories for timber, wildlife, agricultural crops, wetlands and other resources. In almost every case, those relying on information from the inventories have concluded that other information is needed. Plant and animal responses are usually affected by their surroundings. Timber must be characterized by its environment in order to evaluate its availability for harvest. Wildlife inventories usually must combine habitat measures as well as population densities and trends. Before long, it appears that there are many multi resource inventoriesthat each single-resource inventory has now become an inventory of multiple resources. Then pressure is exerted or felt to combine these inventories in order to eliminate waste and overlap. There are good reasons for doing this as well as several reasons for being very cautious in doing so. Unless there is agreement among users that the information derived from a single multi resource inventory meets their needs, the inventory may create more problems than it solves. Putting many things together does not necessarily save money. Larger field crews may be needed, and measurement times may be greatly increased. What follows are some guidelines for a multi resource inventory based on my experience with several single-purpose inventories and some efforts to expand those inventories to multi resource inventories. Sample Unit Size _ _ _ _ _ __ I shall start the discussion with the sampling unit. I know it's desirable to define a sampling frame that consists of a list of all sampling units so that the usual approach of sample selection and use of finite population theory is easily applied. It's not that easy with multi resources-although it's been tried. Some have tried to collect more and more information on the same plot-a plot (sampling unit) that was originally designed for collection of data on one resource. I believe that sampling units must be flexible so that they can be optimized for collection of data on many resources. I believe a useful procedure is to treat a sampling unit as the list of data collected with reference to a point in space and time. Timber information might be collected, in part, by a Ipaper presented at the North American Science Symposium: Toward a Unified Framework for Inventorying and Monitoring Forest Ecosystem Resources, Guadalajara, Mexico, November 1-6, 1998. 2W.E. Frayer is Dean, School of Forestry and Wood Products, Michigan Technological University, Houghton, MI, U.S.A. Phone: (906) 487-3604; Fax: (906) 487-2915; e-mail: wefrayer@mtu.edu USDA Forest Service Proceedings RMRS-P-12. 1999 point sample (probability proportional to basal area). Habitat data may be collected by delineating the boundaries of the habitat the point is located within. Wetland data could be a map of the wetland the point falls within, a description of the type of wetland, distances to other wetlands and any other features needed. Finite population corrections would not be used, because, in essence, there are an infinite number of points contained within the population. This could be considered analogous to sampling with replacement. 1. Recommendation: Use points as sampling units. Many configurations of areas around a point may contribute to variable values for the sampling unit (point). Variation Within and Between Sampling Units As suggested, a sampling unit might be a point with data associated with various resources collected around the point. At any rate, the sampling unit may be complex. Many inventories use a cluster of points/plots as a sampling unit. For some reason, these clusters are often located close together. This may give data gatherers the belief that they're representing a fairly homogeneous area with the data collected. For sampling precision, however, there are very good arguments for maximizing the variation within the clusters and minimizing variation between clusters. The precision of population estimates is strongly dependent on the variation between sampling units-this being a strong argument in favor of absorbing as much variation as possible within the units. It often takes considerable travel time to locate the sampling unit. Why, then, confine the area of the cluster on which measurements are made to an area as small as an acre or a hectare? Spreading the measurements out over a larger area will, on the average, introduce more variability into the cluster and less variability between clusters. In some cases, transects may prove highly useful. 2. Recommendation: Design sampling units such that variation is large within units and small between units. Movement of Sampling Units _ __ The concern here is similar to that of the previous discussion. Some inventory procedures have included the practice of moving sampling units (or portions of sampling units) so that they have minimal variability-as in moving all parts of a cluster into a single vegetative type. This not only is counterproductive to obtaining precise estimates-it can also be a real procedural problem in later inventories when the sampling unit may again straddle vegetative conditions that have changed since the previous inventory. 3. Recommendation: Do not move sampling units. 163 Allocation of Sampling Units _ __ ;:.: Many designs have been tried for sampling single resources and multiple resources. Sampling with partial replacement (SPR) is one of the more complex designs used in forestry (Ware and Cunia, 1962; Bickford et aI., 1963). With most inventories there is an important component ofmonitoring change as well as assessing current status and perhaps forecasting future status under various assumptions. Any time that sampling units are chosen with unequal probability or a complex design like SPR is used, there is a danger that the personnel who try to replicate the procedure for the subsequent inventory will not follow the same, exact procedure used before. Another potential hazard is the change ofthings over time. What was originally an optimum design based on varying probabilities may be far from optimum unless the probabilities can be redefined-regardless of the design used. For example, stratified sampling can become quite laborious to handle if stratum boundaries keep changing. For these reasons, when possible, I believe it best to "keep it simple" and use proportional allocation when there is not a distinct, overriding disadvantage. 4. Recommendation: Use proportional allocation. Sampling Designs _ _ _ _ _ __ As indicated earlier, complex designs like SPR have been used in forest sampling. Multi stage designs up to four stages have been attempted (Schreuder et al. 1992b). A three-phase design using high-altitude photography for an unsupervised classification, low-altitude photography for a more derailed stratification and ground plots for measurement was shown to be advantageous when compared to double sampling for stratification with low-altitude photography and ground plots (Kent et aI., 1979). Stratified sampling with either known or estimated stratum sizes has stood the test of time. The procedure when estimating stratum sizes generally uses some type of remotely sensed data (be it satellite imagery, or aerial photography). Known stratum sizes may be delineated by political boundaries (states, counties), physiographic provinces, or any land classification that proves useful. Multi resource inventories conducted by US Forest Service Experiment Stations all use a form of stratified sampling with estimated stratum weights. Wetland inventories conducted by the US Fish and Wildlife Service use stratified sampling with strata formed by cross sections of political boundaries and physiographic subdivisions. 5. Recommendation: Use designs/procedures that are as simple as possible. Random vs. Systematic Samples _ _ _ _ _ _ _ _ __ Sample points could be located systematically or at random. If located systematically, caution should be used to avoid the potential bias of that approach. For example, you wouldn't want to locate sampling units on a systematic grid if some of the data you want to collect are located on a grid as well. An example of the potential danger would be school lands or railroad holdings in the western United States, both 164 of which were originally laid out in a systematic grid. Other examples of potential problems are roads, fence lines, canals and other features located in periodic fashion. If transects are used, there may be certain instances in which measurements might be more efficient if not equally spaced. 6. Recommendation: Use a systematic design, but with caution. Sampling Over Time _ _ _ _ __ Inventories of natural resources are usually carried out to monitor status and trends. Many designs were adopted because of their precision in estimating status (estimating current values of population parameters). Sampling unit design has often been dependent primarily on how well it could be used to collect data on current values. Point sampling for timber volume (which samples trees with probability proportional to basal area is such an example). If the trends are of much importance, it may make sense to use a sampling design that is optimum for estimating changes. The solution is fairly simple. Ware and Cunia (1962) showed that a complete remeasurement design is optimum for estimating change if the measurements on successive occasions are correlated. Thus, regardless of the exact design used, it is highly recommended that all sampling units be remeasured. 7. Recommendation: Remeasure all sampling units. Changes in Strata Over Time _ __ If stratification is useful-and it generally is as discussed earlier-there must be some acceptable procedure for handling the problem of changing stratum boundaries over time. If working with complete stratification and complete remeasurement of sampling units, the units can be characterized by the stratum at the first or second occasion. For additivity with results from the previous inventory, the original stratification can be used. Post stratification (using the strata as delineated at the second occasion) will provide better correlation with current status and would generally be more precise for estimates of trends as well as status. If the stratum sizes are estimated from a primary sample, the ground plots must be stratified in the same fashion and should truly represent a subsample of the primary sample. In this case also, there are different estimates available depending on the stratification used for the estimation. In this case, too, the most recent stratification is recommended. 8. Recommendation: Use the most recent stratification for all estimates. Future Status _ _ _ _ _ _ _ __ Information on change has assumed more importance in recent years. Presumably, there is reasonably good information on current status and attention has turned to change and how to manage it. Change estimates are often based on average annual change between inventories. Projections are another estimate of change and are becoming more common and of vital use in soliciting funding for affecting future status. Agricultural and forestry organizations have USDA Forest Service Proceedings RMRS-P-12. 1999 routinely dealt with projections. It has been common to project 10, 20, even 50 years into the future. Others have only recently attempted this. The latest status and trends study of U.S. Wetlands used a Markov process for estimation of future status and trends. Without some indication of future status, it was feared that there was little information on which to base federal policies and programs. Given that projections are useful, they should be made with caution. Monte Carlo studies andjackknife estimates have indicated that rates of change used in projection procedures must be very precise for the projections to have any validity at all even for only a few years into the future. 9. Recommendation: Projections should be made with caution. In Summary The recommendations made are based on the experiences of the author. Some individuals may have a different opinion on one or more of these. I believe they form a reasonable set of guidelines, especially for those with very little personal experience who find that they must design a multi resource inventory. When in that situation, an additional recommendation is in order: 10. Seek advice from those with experience! The recommendations given in the paper are summarized here: 1. Recommendation: Use points to derme sampling units. Many configurations of areas around a point may contribute to variable values for the sampling unit. USDA Forest Service Proceedings RMRS-P-12. 1999 2. Recommendation: Design sampling units such that variation is large within units and small between units. 3. Recommendation: Do not move sampling units. 4. Recommendation: Use proportional allocation. 5. Recommendation: Use designs/procedures that are as simple as possible. 6. Recommendation: Use a systematic design, but with caution. 7. Recommendation: Remeasure all sampling units. 8. Recommendation: Use the most recent stratification for all estimates. 9. Recommendation: Projections should be made with caution. 10. Seek advice from those with experience! Literature Cited Bickford, C. A., C. E. Mayer, and K. D. Ware. 1963. An efficient sampling design for forest inventory: the Northeastern forest resurvey. J. For. 61:826-833. Kent, B., D. Johnston, and W. E. Frayer. 1979. The applications of three-phase sampling for stratification to multi-resource inventories. In: Proceedings, Forest Resource Inventories. Colorado State University. pp. 993-1000. Schreuder, H. T., V. J. LaBau, andJ.W. Hazard. 1992b. The Alaska four-phase sampling design. Submitted to Photogramm. Eng. Rem.Sens. Ware, K. D., and T. Cunia. 1962. Continuous forest inventory with partial replacement of samples. Forest Science Monograph No.3. 165