Raymond L. Czaplewski

Design and Analysis for Thematic Map Accuracy Assessment: Fundamental Principles

Abstract Before being used in scientific investigations and policy decisions, thematic maps constructed from remotely sensed data should be subjected to a statistically rigorous accuracy assessment. The three basic components of an accuracy assessment are: 1) the sampling design used to select the reference sample; 2) the response design used to obtain the reference land-cover classification for each sampling unit; and 3) the estimation and analysis procedures. We discuss options available for each of these components. A statistically rigorous assessment requires both a probability sampling design and statistically consistent estimators of accuracy parameters, along with a response design determined in accordance with features of the mapping and classification process such as the land-cover classification scheme, minimum mapping unit, and spatial scale of the mapping.

Adaptive Monitoring Design for Ecosystem Management

Adaptive management of ecosystems (e.g., Holling 1978, Walters 1986, Everett et al. 1994, Grumbine 1994, Yaffee 1994, Gunderson et al. 1995, Frentz et al. 1995, Montgomery et al. 1995) structures a system in which monitoring iteratively improves the knowledge base and helps refine management plans. This adaptive approach acknowledges that action is necessary or appropriate with imperfect knowledge (Raiffa 1968, Walters 1986, Everett et al. 1994, USDA/FS and DOI/BLM 1994b) and that initial actions can be refined as more information becomes available. Imperfect knowledge is the case in ecosystem management particularly when the plan includes the management of complex ecological patterns and processes over large areas and long periods of time. In such cases, we suggest that an adaptive approach should apply not only to the management of the ecosystem, but also to the design of the monitoring program.

Spatial cross-correlation of undisturbed, natural shortleaf pine stands in northern Georgia

In this study a cross-correlation statistic is used to analyse the spatial relationship among stand characteristics of natural, undisturbed shortleaf pine stands sampled during 1961–72 and 1972–82 in northern Georgia. Stand characteristics included stand age, site index, tree density, hardwood competition, and mortality. In each time period, the spatial cross-correlation statistic was used to construct cross-correlograms and cumulative cross-correlograms for all significant pairwise combination of stand characteristics. Both the cross-correlograms and cumulative cross-correlograms identified small-scale clustering and weak directional gradients for different stand characteristics in each time period. The cumulative cross-correlograms, which are based on inverse distance weighting were more sensitive in detecting small-scale clustering than the cross-correlograms based on a 0–1 weighting. Further analysis suggested that the significant cross-correlation observed among basal area growth and other stand characteristics were due, in a large part, on a subset of sample plots located in the northern part of the state, rather than regional or broad-scale variation as first thought. The ability to analyse the spatial relationship between two or more response surfaces should provide valuable insight in the development of ecosystem level models and assist decision makers in formulating pertinent policy on intelligent multiresource management.

Long-term strategy for the statistical design of a forest health monitoring system

A conceptual framework is given for a broad-scale survey of forest health that accomplishes three objectives: generate descriptive statistics; detect changes in such statistics; and simplify analytical inferences that identify, and possibly establish cause-effect relationships. Our paper discusses the development of sampling schemes to satisfy these three objectives, but without any design restrictions implied by existing sample surveys. A general vision of a desirable future system will increase chances that short-term decisions will lead to better environmental monitoring systems in the long term.

Accuracy Assessment of Maps of Forest Condition

No thematic map is perfect. Some pixels or polygons are not accurately classified, no matter how well the map is crafted. Therefore, thematic maps need metadata that sufficiently characterize the nature and degree of these imperfections. To decision-makers, an accuracy assessment helps judge the risks of using imperfect geospatial data. To analysts, an accuracy assessment helps describe the reliability of the map for geospatial analyses and modeling, and the distribution of different types of “true” land cover within each mapped category. To producers of thematic maps, an accuracy assessment measures the degree of technical success for alternative algorithms or techniques. To project managers, an accuracy assessment helps determine contract compliance or measure performance of technical staff.

Evaluation of weighted regression and sample size in developing a taper model for loblolly pine

A stem profile model, fit using pseudo-likelihood weighted regression, was used to estimate merchantable volume of loblolly pine (Pinus taeda L.) in the southeast. The weighted regression increased model fit marginally, but did not substantially increase model performance. In all cases, the unweighted regression models performed as well as the weighted regression models, even for very small sample sizes.

Combining Accuracy Assessment of Land-Cover Maps with Environmental Monitoring Programs

A scientifically valid accuracy assessment of a large-area, land-cover map is expensive. Environmental monitoring programs offer a potential source of data to partially defray the cost of accuracy assessment while still maintaining the statistical validity. In this article, three general strategies for combining accuracy assessment and environmental monitoring protocols are described. These strategies range from a fully integrated accuracy assessment and environmental monitoring protocol, to one in which the protocols operate nearly independently. For all three strategies, features critical to using monitoring data for accuracy assessment include compatibility of the land-cover classification schemes, precisely co-registered sample data, and spatial and temporal compatibility of the map and reference data. Two monitoring programs, the National Resources Inventory (NRI) and the Forest Inventory and Monitoring (FIM), are used to illustrate important features for implementing a combined protocol.

Identifying changes in tree form for harvested ponderosa pine in the Black Hills

Recent underestimates of total volume for timber sales in the Black Hills National Forest prompted analysis of two felled ponderosa pine (Pinus ponderosa Laws.) data sets that were collected approximately 10 years apart. Though neither data set collected was a representative sample of the Black Hills, both were similar in terms of diameter at breast height and total height. We investigated several methods for assessing differences in tree form and applied them to these two data sets. Under the assumption that these two data sets were representative of harvested trees in the Black Hills (which may be incorrect), we concluded that the average tree form of harvested ponderosa pine has changed significantly in the last 10 years. This conclusion highlights the importance of using representative data in model building.