Plot-GEM Your personal assistant for exploration of sampling alternatives
Example #4: We're thinking about stratifying our inventory, but which strata should we use? In 2000, Region 1 (R1) conducted a pilot study in the Idaho Panhandle National Forest. The study area contained 59 existing FIA primary sampling units (PSU's). R1 intensified this grid such that there were 400 PSU's on a regular grid in the study area. R1 had divided the study area into strata that they defined using 14 cover types, two size classes (<9" and >9"), and three density classes (<25%, 25-65%, and >65%). Each subplot was assigned to a stratum. One of the questions R1 wished to answer via the pilot study was "For which strata can we make accurate estimates of the mean of forest attributes?" Plot-GEM was used to generate the graphs in Figures 5, 6, and 7. These three figures depict the percent error graph and 95% confidence interval graph for the estimate of the mean of attribute live basal area per acre in three different strata. The first stratum was defined by cover type = Douglas Fir, size >9", and density 25-65%. Figure 5 shows that mean live basal area in this stratum can be estimated accurately. Even at the sampling intensity of 59 PSU's (existing FIA gird), the estimate of mean live basal area in this stratum is accurate to within ±18%. Figure 5 Percent error graph and 95% confidence interval graph for the estimate of mean live basal area per acre in the stratum defined by cover type = Douglas Fir, size >9", and density 25-65% (df_2_2). The second stratum was defined by cover type = Lodgepole Pine, size >9", and density 25-65%. Figure 6 shows that mean live basal area in this stratum can be estimated accurately with a larger sampling intensity. At the sampling intensity of 59 PSU's (existing FIA gird), the estimate of mean live basal area per acre in this stratum is accurate to within ±29.4%. However, intensify the FIA grid to 150 PSU's will increase the accuracy of the estimate of mean live basal area per acre, reducing the percent error to ±19.6%. Figure 6 Percent error graph and 95% confidence interval graph for the estimate of mean live basal area per acre in the stratum defined by cover type = Lodgepole Pine, size >9", and density 25-65% (lp_2_2). The third stratum was defined by cover type = Mesic Conifer, size <9", and density 25-65%. Figure 7 shows that mean live basal area per acre in this stratum cannot be estimated accurately. Even at a sampling intensity of 400 PSU's, mean live basal area per acre in this stratum can only be estimated to within ±21.5%. Figure 7 Percent error graph and 95% confidence interval graph for the estimate of mean live basal area per acre in the stratum defined by cover type = Mesic Conifer, size <9", and density 25-65% (mescf_1_2). Often, the ability to estimate attributes in strata depends solely on the number of subplots located within the strata. However, that was not the case with this example. There were 86 subplots located in the stratum defined by cover type = Douglas Fir, size >9", and density 25-65% (df_2_2). There were 28 subplots located in stratum defined by cover type = Lodgepole Pine, size >9", and density 25-65% (lp_2_2). There were 39 subplots located in stratum defined by cover type = Mesic Conifer, size <9", and density 25-65% (lp_2_2). Even though more subplots were located in the third stratum than there were in the second stratum, mean live basal area per acre could be more accurately estimated in the second stratum. This example illustrates that the type of stratum can significantly influences the accuracy of forest attribute estimation. In this way R1 used Plot-GEM to determine (a) for which strata attributes could be accurately estimated using the existing FIA grid, (b) for which strata attributes could be accurately estimated if the FIA grid was intensified, and (c) for which strata attributes could never be accurately estimated using a grid-based sampling scheme.

|Return to index|