In this paper, we conduct multivariate analyses of similarity amongst lidar instruments and canopy models by exploiting the similarities of binned canopy height profiles to community datasets in that we treat profile bins as “species” to calculate distance matrices between sample units. Canopy profiles were derived from lidar data using the MacArthur–Horn transformation and from field data using a model we developed that uses two sets of allometric equations describing crown shape and tree height and a third from raw field data. We conducted a statistical comparison of seven asynchronous relative vegetation profiles (RVP) derived from different methodologies between the years of 2005 and 2012. We compared three airborne lidar datasets, three modeled profiles from field data, and one terrestrial lidar dataset using pairwise Mantel tests, multi-response permutation procedure (MRPP) and a permutation-based comparison using a similarity index for within plot comparisons. We used the results of MRPP to determine possible drivers of poor agreement between instruments and models and found a moderate relationship between within-plot variability (observed delta) and estimated above ground biomass (r2=0.273, p<0.05), which we attribute to poor model performance on low density/low biomass plots. In addition, correlation analysis of height metrics derived from RVPs resulted in weak correlations at low height percentiles and strong correlations at higher percentiles. Overall, we identified general similarity between lidar profiles using MRPP (lidar only A=0.202, p<0.001), but poorer agreement between lidar and modeled profiles (all profiles A=0.076, p<0.001). We attribute some of these differences to selection of canopy profile models and to approaches used for accounting for canopy occlusion in lidar transformations. In addition, we discuss the possibility of using Mantel tests to estimate temporal scales of vertical structure change in La Selva Biological Station, Costa Rica.
