OPTIMIZATION OF AN ECOLOGICAL MODEL WITH THE MARQUARDT ALGORITHM

Optimization of parameters in ecological models can present many challenges, including multiple dependent variables on which to calculate fit, slowly converging search algorithms, nonanalytical derivatives, parameters constrained by their physical meaning in the model to a given sign or range, and excessively long run times for optimization calculations. The Marquardt algorithm has been restructured and an appropriate figure-of-merit function formulated to address these problems. As an example, a soil nitrogen and carbon transformation simulation model (NCSOIL) may be optimized for two and three unknown parameters within ten iterations or less, based on initial guesses of parameter values in error by an order of magnitude. © 1990.