Journal
SCANDINAVIAN JOURNAL OF FOREST RESEARCH
Volume 36, Issue 1, Pages 43-54Publisher
TAYLOR & FRANCIS AS
DOI: 10.1080/02827581.2020.1852309
Keywords
Finite mixture model; lasso; geographically weighted regression; spatial stationarity; design based; post-stratification
Categories
Ask authors/readers for more resources
This study addresses the issue of spatial heterogeneity and prediction accuracy improvement in design-based model assisted inference from forest inventory data. Through examples, it demonstrates how to obtain an assisting model, test for spatial stationarity in regression coefficients, and identify spatial model strata for post-stratification.
In design-based model assisted inference from data gathered in a large area forest inventory under a probability sampling design, one should anticipate spatial heterogeneity in the regression coefficients of an assisting model. The consequence of such heterogeneity is that a global estimate of a root mean squared error (RMSE) becomes unsuited for local predictions. With data from the Danish National Forest Inventory, we demonstrate how to: obtain an assisting model with the lasso method; test for spatial stationarity in regression coefficients of the assisting model; and identify spatial model strata for a post-stratification with either a finite mixture modeling or a lasso spatial clustered coefficients method. Spatial model strata apply to any domain and small area estimation problem without the need for complex modeling when domains or small area changes with shifting user needs. One should not a priori expect a spatial model stratification to improve design-based population and strata estimates of precision, but the reliability of domain and small area RMSEs will improve in presence of statistically significant spatial model strata.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available