Background/Question/Methods Understanding the patterns, causes, and consequences of disturbance-regeneration dynamics in tropical forests has been difficult to achieve due to difficulties in reconstructing stand dynamics caused by the lack of annual tree rings and the sparseness of long-term observations in most tropical forests. In the tropical montane rainforests of Hawai`i characterizing plant community response to disturbance is further complicated by the lack of a guild of fast-growing, gap-colonizing species to indicate areas of recent disturbance. At 4000 feet elevation in the Laupahoehoe Experimental Tropical Forest on the Island of Hawai`i we used direct multi-scale ordination (mso) and spatial variance partitioning analysis to determine whether the montane rainforest plant community shows any consistent species compositional variability associated with signs of disturbance and whether there is underlying spatial autocorrelation not explained by other gradients that may indicate the spatial extent of cryptic disturbance patches. To achieve the required large number of observations at many spatial lag distances we used a 980M long block transect of rhythmically arranged sampling quadrats. Constrained ordination provided a multivariate model of how observed environmental gradients are explaining variation in species composition. At each spatial lag the unfitted multivariate residuals matrix (the “unexplained variance”) was compared to the matrix of actual spatial distances with a Mantel permutation test to show unaccounted for spatial autocorrelation in the residuals of the direct ordination.
Results/Conclusions

Indirect non-metric multidimensional scaling (NMS) ordination suggested three major gradients affecting plant community composition in the montane rainforest: 1) substrate age, 2) tree fall disturbance, indicated by the presence of woody debris and 3) surface hydrology. Most species had a neutral relationship to the underlying gradients but several were associated with tree fall disturbance including hapu`u tree ferns (Cibotium glaucum and C. menziesii), akala (Rubus hawaiiensis) and ala ala wai nui (Peperomia hypoleuca and P. cookiana). Variogram partitioning of the variances from direct ordination showed that most of the variance was unexplained. Mantel tests between spatial distance and the ordination residuals showed significant spatial autocorrelation over distances of 0-10 meters and also indicated we were underestimating errors in the multivariate model by 0.4% due to spatial autocorrelation. Plant community composition in the montane rainforests at Laupahoehoe is affected by soil productivity, tree fall disturbance and disturbance by surface water flow but spatially-explicit community assembly processes are also equally important.