Sensitivity analysis and stability patterns of two-species pest models using artificial neural networks

62) PARK, Y-S., J. RABINOVICH AND S. LEK

ECOLOGICAL MODELLING

Año: 2007 vol. 204 p. 427 - 438

0304-3800

Sensitivity analysis is a critical step in mathematical modelling of ecological processes andit provides an idea of the response of the model dynamics to a variation in the values of someparameters. In analytic models, there are standard mathematical techniques for carryingout sensitivity analyses, but this is not so with simulation models, mainly due to the factthat their behaviour usually depends upon the interaction among different parameters, andso sensitivity analysis has to be carried out for all combinations of all parameters of interest.In this study, we explored the use of artificial neural networks (ANN) for sensitivity analysisof simulation models, as applied to simulations models of two-species pest populations:the parasitoid–host system Nezara viridula–Trichopoda giacomellii, N. viridula being a pest ofsoybean and the Sirex noctilio–Pinus radiata system, S. noctilio being a pest of pine plantations.We compare the ANN sensitivity analysis results with the ones of the Classification Trees(CT), Sobol and the stepwise multiple regression with standardized partial regression coefficients(SMR). The sensitivity analyses were carried out evaluating the simulations models’parameters effect on the stability behaviour of the simulation models. The ANN sensitivityanalysis produced the same (or superior) results as the other two techniques (CT, Sobol andSMR), but showed additional advantages similar to those offered by sensitivity analyses ofanalytic models: partial derivatives were calculated to determine the contribution of eachparameter of the simulation models to their stability behaviour. We conclude that ANNis adequate for simulation modelling sensitivity analysis with the additional advantage ofevaluating the contribution of model parameters to the model’s behaviour. Although, weused only two-species pest systems as an example, this approach may be applied in wideareas of pest management and population dynamics studies.