An inverse analysis of a matrix population model using a genetic algorithm

Accurate estimation of demographic parameters of tropical tree population models can be difficult due to low mortality rates coupled with typically short observation durations. In this study, we use a Genetic Algorithm (GA) for inverse parameter estimation of a tropical palm (Mauritia flexuosa) matrix population model. The palm matrix model, with six size classes above 1 m height, simulates a density-dependent mono-dominant population. The population was sampled during 1994 through 1996, and is believed to be at steady-state. The previously published parameter values poorly predict the observed steady-state size class distribution. We found that GA optimization led to greatly improved fits with mean errors of less than one individual per size class. However, repeating the GA optimization 15 times demonstrated a lack of consistency in the magnitudes of optimal demographic parameters, with some parameters far outside the range of estimates from five measurement plots. An additional set of GA optimizations, constrained to keep 13 parameters within the plot-to-plot variation, also had excellent fits, but was much more consistent. This consistent pattern demonstrates that the observed size class distribution is a plausible result of the hypothesized model and the parameter space bounded by measurements. The pattern of optimal parameter estimates in the constrained GA optimization set supports the hypothesis that juvenile palms (6–20 m height) grow rapidly into the reproductive size classes, and that this rapid growth was underestimated in the field sampling.

► We use a Genetic Algorithm to evaluate parameter fits for a palm matrix population model.
► We examine how growth and survival parameters influence the size class distribution of a palm population model.
► Inverse modeling can be used to calibrate population models