Abstract: Based on the aerial-borne small footprint LiDAR point cloud and 73 sample plots from field inventory, this paper sets the subtropical secondary forests as a research subject. First, the methods of principle component analysis (PCA), stepwise regression and bayesian modeling averaging (BMA) were applied to optimize the extraction of LiDAR-derived metrics; second, the optimized models were used to estimate each forest parameter and then the accuracy evaluation was performed; finally, the volume information was up-scaled to map its spatial distribution. The results demonstrated that the optimized LiDAR-derived metrics selected by PCA were average height (hmean), 60% canopy return density (d6 ) and the coefficiency of height variation (hcv), and these metrics were also selected by stepwise regression and BMA. The stepwise regression method fitted the best model (R2 was 0.39 -0.84), while BMA(R2 was 0.32 - 0.77) and PCA (R2 was 0.26 - 0.74) performed a little poor. Among each forest parameters, Lorey's height (R2 was 0.74 -0.84) and dominated height (R2 was 0.73 -0.82) had the highest accuracy, whereas DBH (R2was 0.48 - 0.57) and volume(R2 was 0.46 - 0.55)were a little lower, and stem number (R2 was 0.35 -0.44) and basal area (R2 was 0.29 -0.39) were the lowest.