Abstract
Dual-filter estimators are described and tested for use in the annual estimation for national forest inventories. The dual-filter approach involves the use of a moving widow estimator in the first pass, which is used as input to Theil’s mixed estimator in the second pass. The moving window and dual-filter estimators are tested along with two other estimators in a sampling simulation of 152 simulated populations, which were developed from data collected in 38 states and Puerto Rico by the Forest Inventory and Analysis Program of the USDA Forest Service. The dual-filter estimators are shown to almost always provide some reduction in mean squared error (MSE) relative to the first pass moving window estimators.
Highlights
A rotating panel design for re-measured ground observations has replaced what are known as periodic inventory designs in the United States and elsewhere in an effort to improve the annual estimation of forest attributes
That challenge stems from the common practice in national forest inventory (NFI) to subdivide the land area into estimation units
All of the states and territories covered by the Forest Inventory and Analysis (FIA) in the period from 1995 to 2015 were considered for this study, some states and territories did not yet have enough data in all estimation units to construct a general set of populations for the unit
Summary
A rotating panel design for re-measured ground observations has replaced what are known as periodic inventory designs in the United States and elsewhere in an effort to improve the annual estimation of forest attributes. Where P is the number of panels (and years, given that one panel per year is measured), t is the year of interest, and yi is the observation of the variable of interest in year i This end of period estimator (FIA_EoP) is well-known: it will usually have very low variance because it uses all of the available data, but will be biased in the presence of a non-zero trend. First and foremost, it would be counter to the philosophy and intent of that estimator because the estimator is only intended to provide an estimate for the most recent year of data collection Second it would be an extremely long series to supplement under the assumption used to supplement the other estimators. These seven estimators and their advantages relative to the annual sample mean and to each other are compared in the simulation described below
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have