An Application 



Even a small volume bias can compound rapidly when 

 summing many trees in an inventory. An additional data 

 set was used to examine the bias observed in the volume 

 equations in figures 2 to 5. We used 176 plots of 0.1 

 acre each from the Schell Resource Area of the Ely, NV, 

 BLM District, to compare predictions of the volume 

 equations for the estimator groups. The plots were taken 

 from a systematic grid at a sampling fraction of 0.1 acre 

 for each 6,177 acres. The Schell data are within the sam- 

 ple population for this study and include the same two 

 pinyon and juniper species. 



Table 3 shows a summary for average cubic foot vol- 

 ume per acre (computed 'from the four equations in figs. 

 4 and 5) for the Schell Resource Area. The mean 

 volumes per acre for each estimator look reasonably 

 similar. However, most of the per-acre means, computed 

 from volume equations based on visual data, under- 

 estimate volume when compared to the destructive equa- 

 tions. Overall, a negative bias as high as —9 percent (for 

 pinyon, for BLM and FS2) resulted from using visual 

 segmentation as opposed to destructive segmentation for 

 estimating population means. Results tended to be bet- 

 ter for juniper than for pinyon. 



DISCUSSION 



Should visual segmentation be used in a woodland 

 inventory to estimate volume? This study shows it is a 

 useful technique, but, like any subjective method, it 

 should be used with caution. Estimators tend both to 

 overestimate and underestimate volumes visually, with 



underestimates more common. Juniper fared better than 

 pinyon, perhaps because juniper trees usually have more 

 segments per tree than do pinyon. From the theory 

 behind visual segmentation, more segments per tree give 

 the technique higher probability of success (see equation 

 2). Applications of visual segmentation to real data 

 showed a to —9 percent bias in estimating mean vol- 

 ume per acre. 



Perhaps doing more than one visual estimate per tree 

 for fewer trees would be an effective way to reduce the 

 bias rather than estimating many trees once. Table 4 

 shows the improvement by grouping estimates. The 

 median or mean of three estimates appears to be the 

 best for the Nevada data. 



The theory behind visual segmentation appears sound, 

 based on the good results from estimator FSl. How 

 ever, the less consistent performance of the BLM esti- 

 mator is probably closer to production mode reality. But 

 with proper quality control, visual segmentation is cer- 

 tainly worth consideration. Our more recent experience 

 indicates that intense training, consistent use of the seg- 

 ment poles, and quality control checks on estimators sig- 

 nificantly improves the quality of the visual data. 



We would like to make a final point somewhat 

 unrelated to this study but important for application of 

 visual segmentation to volume equation construction. 

 Even if the bias from visually measuring pinyon-juniper 

 volume was entirely eliminated, the procedure would not 

 be completely satisfying for developing volume equa- 

 tions. This is because the relationship between pinyon- 

 juniper tree volume and easily measured variables is not 

 well understood. 



Table 3.— Summary of 176 forested plots from thie Schell Resource Area of the Ely BLM District. Volumes for each 

 tree were computed using volume equations developed for each estimator group 



Volume per acre Sampling Percent of plots 



Species _ Diameter class BLM FS1 FS2 Destructive error^ having trees^ 



Inches Ft-^ Percent 



Juniper 





3- 6.9 



9.8 



8.0 



7.5 



8.6 



±22 



68 



Juniper 





7-10.9 



37.4 



33.6 



32.6 



35.2 



±22 



65 



Juniper 





11 - 14.9 



52.1 



49.5 



49.1 



50.9 



±19 



57 



Juniper 





15-18.9 



69.7 



69.4 



70.1 



70.3 



±22 



46 



Juniper 





>19 



131.3 



138.5 



142.9 



137.8 



±28 



38 



Juniper 





Total 



300.3 



299.0 



302.2 



302.8 



±17 



89 



Pinyon 





3- 6.9 



32.4 



33.2 



30.0 



31.9 



±21 



78 



Pinyon 





7-10.9 



76.3 



81.6 



74.4 



81.0 



±21 



64 



Pinyon - 





11 - 14.9 



70.4 



77.6 



71.2 



78.6 



±28 



38 



Pinyon 





15-18.9 



55.2 



62.3 



57.5 



64.1 



±38 



19 



Pinyon 





>19 



26.1 



29.8 



27.6 



30.9 



+ 56 



7 



Pinyon 





Total 



260.4 



284.5 



260.7 



286.5 



±18 



84 



Pinyon and 



juniper 



Total 



560.7 



583.5 



562.9 



589.3 



±13 



100 







Volume Bias^ as Compared to the Destructive Equation 







Juniper 





Total 



-1% 



-1% 



0% 





±17 



89 



Pinyon 





Total 



-9% 



-1% 



-9% 





±18 



84 



Pinyon and 



juniper 



Total 



-5% 



-1% 



-4% 





±13 



100 



^The sampling error is a t-statistic confidence interval, expressed as a percent of the sample mean, that contains the population 

 mean unless a 1-in-20 chance occurs in selecting the sample. The errors were the same for all four volume estimates in a diameter 

 class. 



^This refers to the percentage of plots having trees for that diameter class. The volume was set to zero in the variance computa- 

 tion for the plots having no trees in a given diameter class. 



■'Bias is defined as a visual estimate (BLM, FSl, or FS2) minus the destructive estimate, divided by the destructive estimate. 



7 



