Comparison of annual inventory designs using forest inventory and analysis data

• Published in: Forest science Vol. 50; no. 2; pp. 188 - 203
• Main Authors: Arner, S.L, Westfall, J.A, Scott, C.T
• Format: Journal Article
• Language: English
• Published: Bethesda Oxford University Press 01.04.2004
• Subjects: Comparative studies; data analysis; Design of experiments; equations; error analysis; estimation; forest inventory; Forestry; Inventory; mathematical models; research methods; statistical analysis; Vermont; New York; West Virginia; Maine; Massachusetts; New Hampshire; Kentucky; Pennsylvania
• ISSN: 0015-749X; 1938-3738
• DOI: 10.1093/forestscience/50.2.188

Three annual inventory designs, a periodic design, and a periodic measurement with midcycle update design are compared using a population created from 14,754 remeasured Forest Inventory and Analysis plots. Two of the annual designs and the midcycle update design allow updating of plots not measured using sampling with partial replacement procedures. Comparisons are based on root mean square error and estimator bias determined for net volume (m3/hectare) and mean annual net volume change. For the annual designs, both individual year and moving average estimates are investigated. The latter are compared to both the population means of the most recent year used in the average and to population means covering the same period as the estimate. Among annual designs, a rotating panel design, in which an equal portion of the total sample is measured each year without remeasurement until the start of the next measurement cycle, produced the smallest root mean square error for estimates of mean net volume. For multiple-year comparisons, the rotating panel and periodic designs resulted in the smallest root mean square errors; for single-year comparisons, the periodic design resulted in the smallest root mean square error. For mean annual volume change, the smallest root mean square error was produced by the periodic design. Among annual designs, the rotating panel design resulted in the smallest root mean square error for multiple-year comparisons of volume change, while a design allowing annual updates of estimates using generalized least squares resulted in the smallest root mean square error for single-year comparisons.