Tree Species Occurrence Related to Dry Season Duration, Panama Canal Area

Abstract

Woody plant species with free-standing stems >=1 cm diameter were surveyed at 72 locations near the Panama Canal, each with an area <0.5 square kilometer. Sites were arranged to span geological formations and the rainfall gradient; all were within 65 km of one other. Most were low elevation, with just two sites >600 m above sea level, the highest at 888 m. All were in closed-canopy forest, including undisturbed old-growth and secondary stands 60-100 years old. Forty surveys were permanent census plots in which every individual tree was located, measured, and identified: a 50-ha plot at Barro Colorado, a 5.96-ha plot at Fort Sherman, and 38 1-ha plots. At 32 other sites, surveys were one-day inventories in which all tree species were noted until we could find no more, without counting individuals. Nearly every tree was identified, either on the spot or later after comparing against keys or guides and herbarium specimens (Smithsonian Tropical Research Institute and the University of Panama).

The Meteorology and Hydrology Branch of the Panama Canal Authority, Republic of Panama, maintains a network of meteorological stations throughout the Panama Canal Watershed. Daily rainfall at the 46 of those stations having 3-47 years of uninterrupted data since 1960, plus daily records from Barro Colorado Island station, were acquired by Smithsonian Tropical Research Station Physical Monitoring Program.

The mean annual extreme of dry-season moisture at the 47 rainfall gauges was calculated as the cumulative moisture deficit (precipitation minus potential evapotranspiration). This dry season measure was then interpolated at 72 forested sites across the Panama Canal area using an optimized spatial kernel fitted to the results at the 47 gauges.

Eight environmental predictors were used to model of tree distributions: dry season moisture plus soil Al, Ca, Fe, K, P, N, Zn. Soil concentrations were log-transformed, then each predictor was standardized to mean 0 and standard deviation 1. Species occurrences were modeled against the eight factors simultaneously using Gaussian logistic regression. All species were included at once using a hierarchical Bayesian approach (a multilevel regression. The hierarchy consists of a lower level of species-level response parameters beneath an overarching community-wide distribution of those responses. The upper level --- the hyperdistribution and particularly its standard deviation --- measures the variability of responses across a community and thus reveals whether a resource differentiates species. The lower level provides a separate measure for the response of every species to each environmental factor. The hierarchical aspect was critical, preventing over-fitting in rare species but weighting all species according to frequency. Parameters were fitted with a Gibbs sampler based on Metropolis updates, producing credible intervals for all statistics.

These results were published in Condit R, Engelbrecht BMJ, Pino D, Pérez R, Turner BL (2013) Species distributions in response to individual soil nutrients and seasonal drought across a community of tropical trees. Proc Natl Acad Sci USA, 10.1073/pnas.1218042110.