INVESTIGADORES
BLANCO Anibal Manuel
congresos y reuniones científicas
Título:
An artificial neural network approach for modelling weed emergence dynamics: the case of Avena fatua L.
Autor/es:
CHANTRE GUILLERMO; BLANCO ANíBAL M.; LODOVICHI, MARIELA V.; SABBATINI, M. RICARDO; BANDONI J. ALBERTO
Lugar:
Hangzhou
Reunión:
Congreso; 6th International Weed Science Congress; 2012
Resumen:
Weed emergence predictive models are essential tools for the implementation of more efficient and sustainable control tactics in agricultural systems. Avena fatua is worldwide distributed noxious weed which produces severe yield losses in cereal crops. In the semiarid temperate region of Argentina, A. fatua emergence patterns show great variability between years mainly due to a highly unpredictable precipitation regime and a fluctuating thermal environment. Artificial neural networks (ANN) models are known for their capacity to describe highly non-linear relationships among variables thus showing a high potential applicability in ecological systems. The objectives of the present work were to: (i) develop different ANN model architectures based on ecophysiological indexes; (ii) compare ANN predictive accuracy with traditional non-linear regression models. A multilayer perceptron structure with three layers (input, hidden and output layer) and a Bayesian Regularization Backpropagation algorithm were used. Input variables for model development were: hydrothermal-time (èHT) and the combination of thermal-time (èT) and hydro-time (èH) as independent variables. The accumulated proportion of seedling emergence was the output variable. A total of 528 input/output data pairs corresponding to 11 years of data collection were divided into training (82%) and test (18%) subsets. Traditional non-linear regression (NLR) sigmoid shape models were evaluated (Weibull, logistic, general logistic). ANN and NLR models based on èHT showed similar goodness of fit (RMSEtrain = 0.214; r = 0.84) and predictive capability (RMSEtest = 0.177; r = 0.92), irrespective of the number of model parameters. Conversely, an ANN model with èT and èH as input variables and 30 effective parameters gave the best prediction of emergence data (RMSEtest = 0.078; r = 0.98) showing a satisfactory generalization capacity. These results indicate an advantage of ANN models over NLR methods further suggesting its potential applicability in weed management decision support systems.