The influence of LiDAR pulse density and plot size on the accuracy of New Zealand plantation stand volume equations

Background: LiDAR is an established technology that is increasingly being used to characterise spatial variation in important forest metrics such as total stem volume. The cost of forest inventory and LiDAR acquisition are strongly related to the inventory plot size and the LiDAR pulse density, respectively. It would therefore be beneficial to understand how reductions in these variables influence the strength of relationships between LiDAR and stand metrics. Although relatively high pulse densities are required for creating Digital Terrain Models (DTMs), once a DTM has been developed there is scope for reducing pulse density on subsequent flights to estimate stand metrics from LiDAR. This study used an extensive national dataset (for which the DTM had been characterised) obtained within New Zealand's planted forests. Using this dataset, the objective of this research was to investigate how variation in both pulse density and plot size influence the precision of relationships between LiDAR metrics and total stem volume. Methods: LiDAR metrics were thinned to pulse densities ranging from 0.01 to 4 pulses m(-2) across plot sizes ranging from 0.01 to 0.06 ha. For each pulse density/plot size combination regressions between LiDAR mean height and total stem volume were fitted using parameters fixed at values for the unthinned dataset or separately fitted for each pulse density/plot size combination. Results: Using the unthinned dataset (plot size = 0.06 ha; pulse density = 4 pulses m(-2)) the relationship between the mean LiDAR height and total stem volume had a coefficient of determination (R-2) of 0.77. Thinning of the data had little effect on R-2 above plot sizes of 0.03 ha and pulse densities of 0.1 pulses m(-2). As pulse densities decreased below 0.1 pulses m(-2) within plots of less than 0.025 ha, there was a sharp decline in R-2 reaching values as low as 0.48 in plots of 0.01 ha with pulse densities of 0.01 pulses m(-2). Simulations where parameters were fixed yielded almost identical R-2 values to those where they were refitted for each plot size/pulse density combination. The number of pulses per plot integrates the effect of these two factors, with little change in the precision of the volume function until a threshold of 100 pulses per plot was reached. Conclusions: This study showed that the precision of LiDAR-volume equations was relatively insensitive to reductions in pulse density and plot size when an accurate DTM was available. Acquisition of LiDAR information at lower pulse densities is likely to markedly improve the cost efficacy of this information for inventory purposes.