Appropriate search techniques to estimate Weibull function parameters in a Pinus spp. plantation

Abstract

The Weibull function, a continuous probability distribution, is widely used for diameter distribution modelling, in which parameter estimation performance is affected by stand attributes and fitting methods. The Weibull cumulative distribution function is nonlinear, and classical fitting methods may provide a not optimal solution. Invoking the use of artificial intelligence by metaheuristics is reasonable for this optimisation task. Therefore, aimed and compared (1) the metaheuristics genetic algorithm and simulated annealing performance over the moment and percentile methods; (2) the hybrid strategy merging the metaheuristics tested and the percentile method and, (3) the metaheuristics fitness functions under basal area and density errors. A long-term experiment in a Pinus taeda stand subjected to crown thinning was used. According to our findings, all methods have a similar performance, independent of the thinning regimes and age. The pattern of the estimated parameters tends to be acceptable, as b increases over time and c increases after thinning. Overall, our findings suggest that methods based on metaheuristics have a higher precision than classical methods for estimating Weibull parameters. According to the classification test, the methods that involved simulated annealing stood out. The hybrid method involving this metaheuristic also stood out, with accurate estimates. Classical methods showed the poorest performance, and we therefore suggest the use of simulated annealing due to its faster processing time and high-quality solution.