H. T. Schreuder’s research while affiliated with United States Department of Agriculture and other places

22 p. A grid sampling strategy was adopted for broad-scale inventory and monitoring of forest and range vegetation on National Forest System lands in the Pacific Northwest Region, USDA Forest Service. This paper documents the technical details of the adopted design and discusses alternative sampling designs that were considered. A less technical description of the selected design will be given elsewhere. The grid consists of a regular, square spacing with 5.47 kilometers (3.4 mi) between grid points. The primary sampling unit (PSU), established at each grid sampling point, consists of a circular, 1-hectare (2.47-acre) plot. The PSU is subsampled with a set of different-sized fixed-area subplots, as well as line transects, to assess all components of vegetation. The design is flexible and can be used with many types of maps. The theory of point and change estimation is described, as well

Inventory data are often used to estimate the area of the land base that is classified as a specific condition class. Examples include areas classified as old-growth forest, private ownership, or suitable habitat for a given species. Many inventory programs rely on classification algorithms of varying complexity to determine condition class. These algorithms can be simple decision trees applied in the field or computer calculations applied on a field data recorder or after the data are collected. The advantages to using these algorithms are consistent classification of the condition class, reduced crew training, and the ability to define new condition classes after the data are collected, which will be referred to as postclassification. We discuss three types of the errors that can occur when these types of algorithms are employed and quantify the potential for error with examples. The examples are substantial oversimplifications of the true problem, but they show how difficult it is to determine anything but the most general condition classes using plot data alone. A discussion of how condition class is scale dependent and some general guidelines and recommendations are given.

Assuming volume equations with multiplicative errors, we derive simple conditions for determining when measurement error in total height is large enough that only using tree diameter, rather than both diameter and height, is more reliable for predicting tree volumes. Based on data for different tree species of excurrent form, we conclude that measurement errors up to ±40% of the true height can be tolerated before inclusion of estimated height in volume prediction is no longer warranted.

Equations for predicting tree volume are often developed using data collected either by a model-based method such as purposive sampling or by stratified random sampling so that an "adequate" number of trees from each diameter class are sampled across the range of classes expected in populations of interest. Such equations are then used together with a design-based (probabilistic) sample such as variable radius plot sampling from a specific population to generate estimates of total volume. The probabilities of selection of the sample trees used in developing the volume equation are ignored, may not be known, or may not be appropriate for populations to which the equation are applied. Less biased and more efficient estimates of the population volume can be generated by using known frequencies or estimated frequencies of the diameter classes in the population from the probabilistic sample used for estimating total volume in the population. These frequencies are used as weighting factors in the construction of population-specific volume equations. We show a reduction in bias and increased efficiency in a simulation study for several forest populations with strong linear relationships between variables and reasonably well known error structure. A model-based sampling procedure called pscx sampling or a large-sample extension thereof is used to select sample trees for volume equations. Such bias reduction did not happen for other populations with weak linear relationships and unknown error structure.

Fixed-area plot sampling, variable-radius plot sampling, and horizontal-line plot sampling performed similarly in estimating competition from adjacent trees if only plots were considered that had an adequate number of sample trees to compute the estimated indices. The percentage of plots for which the indices can be computed tends to be largest for variable-radius plot sampling, then line sampling, and last for fixed-area plot sampling. Variable-radius plot sampling performs about as well as fixed-area plot sampling in computing the competition measures used in the distance-dependent tree models evaluated in this analysis.

Poisson (3P) sampling is a commonly used method for generating estimates of timber volume. The usual estimator employed is the adjusted estimator, Y hata. The efficiency of this estimator can be greatly influenced by the presence of outliers. We formalize such a realistic situation for high-value timber estimation for which Y hata is inefficient. Here, yi approx beta xi for all but a few units in a population for which yi is large and xi very small. This situation can occur when estimating the net volume of high-value standing timber, such as that found in the Pacific Northwest region of the United States. A generalized regression estimator and an approximate Srivastava estimator are not affected by such data points. Simulations on a small population illustrate these ideas.

We present the statistical theory of inventory and monitoring from a probabilistic point of view. We start with the basics and show the interrelationships between designs and estimators illustrating the methods with a small artificial population as well as with a mapped realistic population. For such applications, useful open source software is given in Appendix 4. Various sources of ancillary information are described and applications of the sampling strategies are discussed. Classical and bootstrap variance estimators are discussed also. Numerous problems with solutions are given, often based on the experiences of the authors. Key additional references are cited as needed or desired.

You may order additional copies of this publication by sending your mailing information in label form through one of the following media. Please specify the publication title and number. Abstract Schreuder, H. T.; Williams, M. S.; Aguirre-Bravo, C.; Patterson, P. L. 2003. Statistical strategy for inventorying and monitoring the ecosystem resources of the Mexican States of Jalisco and Colima at multiple scales and resolution levels. Gen. Tech. Rep. RMRS-GTR-107. Ogden, UT: U.S. Depart-ment of Agriculture, Forest Service, Rocky Mountain Research Station. 15 p. The sampling strategy is presented for the initial phase of the natural resources pilot project in the Mexican States of Jalisco and Colima. The sampling design used is ground-based cluster sampling with post-stratification based on Landsat Thematic Mapper imagery. The data collected will serve as a basis for additional data collection, mapping, and spatial modeling products to be described in detail in later documents. Estimation described in this document will be primarily useful for strategic planning at the state and national levels. Because it is a pilot study, the approach for the actual statewide and national inventories is likely to be modified as results are obtained and studied.

Truth for vegetation cover percent and type is obtained from very large-scale photography (VLSP), stand structure as measured by size classes, and vegetation types from a combination of VLSP and ground sampling. We recommend using the Kappa statistic with bootstrap confidence intervals for overall accuracy, and similarly bootstrap confidence intervals for percent correct for each category and user and producer accuracy. A procedure is given for mapped plots to be assessed as being partially or totally correct. We recommend the use of primary accuracy for management decisions and secondary accuracy for research decisions to distinguish between accuracy desired.