COS 93-1
Testing the accuracy of static snapshot census data to predict long-term species abundance changes at a tropical forest on Barro Colorado Island, Panama
Markov matrix models have a more than 40-year history of use in forest ecology to project changes in tree species abundance and forest dynamics from short-term static census data. Most widely used matrix models are stage-based, which is often heavily parameterized using transition probabilities for each predefined stage. Less frequently, matrix models that calculate replacement probabilities have been constructed, assuming that sub-adult individuals beneath the crown of adults eventually replace adults. However, to the best of our knowledge, their accuracy has never been tested with actual long-term data on forest dynamics, and hence their usefulness in projecting future population trends. Using long-term dataset on forest dynamics from the 50 ha forest dynamics plot on Barro Colorado Island (BCI), Panama, we tested how accurately such matrix models generated from data from the first BCI census (1982) projected relative species abundance observed in the 2010 census. We projected the abundances of 10 common canopy and midstory shade-tolerant tree species by calculating replacement probabilities based on the relative frequencies of sapling species (1-8 cm dbh) beneath adult species (dbh > 30 cm). Each matrix yields an eigenvector of predicted equilibrium abundances for each species, which we compared with observed species abundances.
Results/Conclusions
Over nearly a 30-year period, species abundance of the 10 focal species changed substantially in several species, ranging between -33% and +109%. The accuracy of projected long-term change in population abundance decreased with increasing sapling size class. In 10 out of 11 cases, the change was correctly predicted when the smallest size class was used. However, no one sapling size class generated matrix models that provided accurate predictions for all species. In three species, matrix models projections were accurate from all sapling size classes tested. At the other extreme, there was one species whose abundance in 2010 was not predicted well by any sapling size class. The larger the observed changes, the greater the predicted changes were but there was no relationship between how much population abundance changed and whether projections were accurate or not. Matrix models were fairly accurate predictors of changes in species abundances in the BCI forest when replacement probabilities were calculated with smallest sapling size class. By carefully assessing the size class used and the assumptions that are made, simple matrix models using replacement probabilities may have a previously unrecognized potential in projecting forest tree population abundances.