multiple Corine Land Cover polygons crossed individual segment boundaries causing designation of segments to multiple strata. The Corine Land Cover data were
evaluated as a covariate to define strata based on an agricultural intensity index
(Gallego et al. 1999). Hansen and Wendt tested using United States Geological
Survey National Gap Analysis Program classifications for stratification of United
States Department of Agriculture (USDA) Forest Service’s Forest Inventory Analysis plots in Indiana and Illinois (Hansen and Wendt 2000). They noticed the
increased precision of forest inventory estimates. Dunham et al. further evaluated
stratifications using an automated method based on the 1992 National Land Cover
Database (NLCD) and using visual analysis of photo imagery in western Oregon and
concluded that forest inventory estimation accuracies were similar but the cost was
reduced using the automated land cover–based approach (Dunham et al. 2003).
Likens et al. used land cover classification results from Moderate Resolution Imaging Spectroradiometer data for Forest Inventory Analysis stratification and found
that the results were inferior to those based on the NLCD due to a coarser spatial
resolution (Liknes et al. 2004).
Perry used visual interpretation of Landsat Thematic Mapper data and soil maps
for sub stratification of the NASS Arkansas area frame. A manual reordering
procedure was conducted to group area frame PSUs into substrata, which resulted
in a reduction in sampling variance for major crops when compared with the
traditional NASS serpentine ordering process. Implementation was impractical due
to the statistical expertise and manual labor required to conduct reordering (Perry
2000). Perry and Gentle further developed an automated procedure based on simulated annealing, which was tested in the intensive agricultural land use category of
the Arkansas area frame. Variances were further reduced but implementation
remained “impractical” based on available resources (Perry and Gentle 2000).
McRoberts et al. conducted post-stratification forest area estimation for the states
of Indiana, Iowa, Minnesota, and Missouri based on the 1992 National Land Cover
Data set (Vogelman et al. 2001) classes and observation of forest inventory plots.
They concluded that the 1992 NLCD provided an effective means of poststratification, which resulted in reduced estimated variances for estimators of forest
land area over post-stratification based on visual interpretation (McRoberts et al.
2001; McRoberts and Wendt 2002). Likens et al. further evaluated the use of the
2005 Wisconsin CDL and the 1992 NLCD for post-stratification of Forest Inventory
Analysis plots. The results indicated that the 2005 CDL outperformed the 1992
NLCD for post-stratification (Liknes et al. 2009). These studies indicated that
utilizing geospatial land cover classification data for area frame stratification might
result in improved estimates and significantly reduce stratification cost.
14.2.2 NASS Area Sampling Frames
NASS’s area frame land use stratification divides land area using physical boundaries on the ground (i.e., roads, railroads, and rivers) into broad land-use categories
270
C. G. Boryan and Z. Yang
evaluated as a covariate to define strata based on an agricultural intensity index
(Gallego et al. 1999). Hansen and Wendt tested using United States Geological
Survey National Gap Analysis Program classifications for stratification of United
States Department of Agriculture (USDA) Forest Service’s Forest Inventory Analysis plots in Indiana and Illinois (Hansen and Wendt 2000). They noticed the
increased precision of forest inventory estimates. Dunham et al. further evaluated
stratifications using an automated method based on the 1992 National Land Cover
Database (NLCD) and using visual analysis of photo imagery in western Oregon and
concluded that forest inventory estimation accuracies were similar but the cost was
reduced using the automated land cover–based approach (Dunham et al. 2003).
Likens et al. used land cover classification results from Moderate Resolution Imaging Spectroradiometer data for Forest Inventory Analysis stratification and found
that the results were inferior to those based on the NLCD due to a coarser spatial
resolution (Liknes et al. 2004).
Perry used visual interpretation of Landsat Thematic Mapper data and soil maps
for sub stratification of the NASS Arkansas area frame. A manual reordering
procedure was conducted to group area frame PSUs into substrata, which resulted
in a reduction in sampling variance for major crops when compared with the
traditional NASS serpentine ordering process. Implementation was impractical due
to the statistical expertise and manual labor required to conduct reordering (Perry
2000). Perry and Gentle further developed an automated procedure based on simulated annealing, which was tested in the intensive agricultural land use category of
the Arkansas area frame. Variances were further reduced but implementation
remained “impractical” based on available resources (Perry and Gentle 2000).
McRoberts et al. conducted post-stratification forest area estimation for the states
of Indiana, Iowa, Minnesota, and Missouri based on the 1992 National Land Cover
Data set (Vogelman et al. 2001) classes and observation of forest inventory plots.
They concluded that the 1992 NLCD provided an effective means of poststratification, which resulted in reduced estimated variances for estimators of forest
land area over post-stratification based on visual interpretation (McRoberts et al.
2001; McRoberts and Wendt 2002). Likens et al. further evaluated the use of the
2005 Wisconsin CDL and the 1992 NLCD for post-stratification of Forest Inventory
Analysis plots. The results indicated that the 2005 CDL outperformed the 1992
NLCD for post-stratification (Liknes et al. 2009). These studies indicated that
utilizing geospatial land cover classification data for area frame stratification might
result in improved estimates and significantly reduce stratification cost.
14.2.2 NASS Area Sampling Frames
NASS’s area frame land use stratification divides land area using physical boundaries on the ground (i.e., roads, railroads, and rivers) into broad land-use categories
270
C. G. Boryan and Z. Yang