*3.4. Comparison of the Three Models Predicting FSV*

In the training phase, BBM (Figure 8a) with an *R<sup>2</sup>* = 0.92 is slightly better than VBM (Figure 8c) with an *R2* = 0.91. However, the RMSE = 11.90 m3ha−<sup>1</sup> of the VBM is lower than the RMSE = 12.23 m3ha−<sup>1</sup> of the BBM. The BVBM (Figure 8e) has the highest *R2* = 0.93 and the smallest RMSE = 10.82 m3ha−1. In the testing phase, the BBM (Figure 8b) with an *R<sup>2</sup>* = 0.59 and RMSE = 27.72 m3ha−<sup>1</sup> performed almost the same as VBM (Figure 8d) with an *R<sup>2</sup>* = 0.59 and RMSE = 27.32 m3ha−1. Similarly, the BVBM (Figure 8f) had the best performance with an *R2* = 0.60 and RMSE = 27.05 m3ha−1. Obviously, the BVBM is the optimal model in this study, and its predicted FSV is used as the final estimation result to map the FSV. A summary of the data characteristics of FSV as predicted by the three models is presented in Table 6.

**Figure 7.** (**a**,**c**,**e**) are the distribution of error rate versus mtry; (**b**,**d**,**f**) are the distribution of the error versus ntree; (**a**,**b**) are related to the BBM; (**c**,**d**) are related to the VBM; and (**e**,**f**) are related to the BVBM.


**Table 5.** The best mtry, ntree, and performance of the three models.

**Figure 8.** Comparison of the measured FSV and predicted FSV by the three models. (**a**), BBM in the training phase. (**b**), BBM in the testing phase. (**c**), VBM in the training phase. (**d**), VBM in the testing phase. (**e**), BVBM in the training phase. (**f**), BVBM in the testing phase.


**Table 6.** Characterization of FSV predicted by the three models.
