The ability to predict changes in the abundances of the species in ecological communities is essential for sustainable management, biodiversity conservation, and community restoration.We propose a framework to predict such changes. We test our method, which uses the linear Lotka-Volterra equations (LLVE) as well as other empirical predictors (linear least squares regression, quadratic extrapolation, simple exponential smoothing), against the measured abundances of trees from the long-term 50-ha plot on Barro Colorado Island (BCI), along eight censuses.To obtain the parameters of the LLVE -the intrinsic growth rate r and the carrying capacity K of each species and the interspecific interaction matrix A- we first estimate A through the Maximum Entropy (MaxEnt) method. Next, using A as input, we fit r and K. Then, feeding the LLVE with these parameters, we obtain predicted species trajectories along censuses. Since for this particular community the interspecific interaction coefficients are much smaller than the intraspecific ones, keeping only intraspecific competition is enough to predict the evolution of the abundances of several tree species, i.e. the LLVE reduce to a set of uncoupled logistic equations. However, this simplification is not a requirement of the method. We define P-values to establish when the predicted trajectory for a species is statistically significant; this is crucial in determining the set of species over which a particular predictor can be meaningfully applied.To illustrate a possible application of the method, we present our predictions for the abundances of tree species for the currently underway BCI 2020 census, which provide warnings regarding species that are likely to experience important population loss.