Canadian Forest Service Publications

A Pòlya-urn resampling scheme for estimating precision and confidence intervals under one-stage cluster sampling: application to map classification accuracy and cover-type frequencies. 2004. Magnussen, S.; Stehman, S.V.; Corona, P.; Wulder, M.A. Forest Science 50(6): 810-822.

Year: 2004

Issued by: Pacific Forestry Centre

Catalog ID: 25259

Language: English

Availability: Not available through the CFS (click for more information).

Mark record

Abstract

A Pòlya-urn resampling scheme (PURS) is introduced as an alternative to design-based estimation (DBE) for applications related to map accuracy and estimating cover-type frequencies from land-cover maps. PURS is a conceptually simple but computationally intensive alternative to the complexity of deriving Taylor series approximation of DBE variance and the reliance on asymptotic normality for construction of confidence intervals. The potential advantages of this recent resampling scheme are explored in simulated sampling of actual data from three study sites ranging from 102 to 107 km2. PURS root mean square errors (rmse) of six measures of classification accuracy were slightly lower than Taylor series approximations of DBE rmse. PURS estimated standard errors and coverage rates of confidence intervals (95%) for bias-corrected classified cover-type frequencies were, in most cases, superior to corresponding Taylor series approximations of DBE estimates. Yet estimates of variance and confidence intervals for a cover-type occupying less than 5% of a study site were consistently underestimated by both methods. PURS is adaptable to a wide range of forest resource survey designs.