Modelling Avena fatua seedling emergence: a comparative study between traditional

G. CHANTRE; M. LODOVICHI; A. BLANCO; A. BANDONI; R. SABATTINI; R. LOPéZ; M. VIGNA; R. GIGóN

Viña del Mar

Congreso; ALAM 2011; 2011

ALAM

Avena fatua is an invasive weed of the semiarid temperate region of Argentina. Seedling emergence patterns show great variability between years mainly due to a highly unpredictable precipitation regime and a fluctuating thermal environment. Traditional non-linear regression techniques are unable to give a precise prediction of emergence patterns under semiarid conditions. 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 and obtain an optimal model for seedling emergence prediction; (ii) compare the predictive accuracy of traditional non-linear regression models with the ANN approach. Input variables for model development were: hydrothermal-time (èHT), thermal-time (èT) and hydro-time (èH) indexes. 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. Evaluated ANN model scenarios resulted of varying the number of effective parameters using hyperbolic tangent sigmoid transfer functions and a Bayesian Regularization algorithm. Results showed that an ANN model with èT and èH as independent variables and 30 effective parameters gave the most accurate prediction on independent emergence data (RMSE=0.078). Traditional non-linear models showed a lower predictive capability (RMSE=0.132– 0.204) irrespective of the index used. These results indicate a higher accuracy and generalization performance of the optimal ANN model in comparison to traditional non-linear regression further suggesting the potential applicability of the former in weed management decision support systems.is an invasive weed of the semiarid temperate region of Argentina. Seedling emergence patterns show great variability between years mainly due to a highly unpredictable precipitation regime and a fluctuating thermal environment. Traditional non-linear regression techniques are unable to give a precise prediction of emergence patterns under semiarid conditions. 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 and obtain an optimal model for seedling emergence prediction; (ii) compare the predictive accuracy of traditional non-linear regression models with the ANN approach. Input variables for model development were: hydrothermal-time (èHT), thermal-time (èT) and hydro-time (èH) indexes. 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. Evaluated ANN model scenarios resulted of varying the number of effective parameters using hyperbolic tangent sigmoid transfer functions and a Bayesian Regularization algorithm. Results showed that an ANN model with èT and èH as independent variables and 30 effective parameters gave the most accurate prediction on independent emergence data (RMSE=0.078). Traditional non-linear models showed a lower predictive capability (RMSE=0.132– 0.204) irrespective of the index used. These results indicate a higher accuracy and generalization performance of the optimal ANN model in comparison to traditional non-linear regression further suggesting the potential applicability of the former in weed management decision support systems.èHT), thermal-time (èT) and hydro-time (èH) indexes. 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. Evaluated ANN model scenarios resulted of varying the number of effective parameters using hyperbolic tangent sigmoid transfer functions and a Bayesian Regularization algorithm. Results showed that an ANN model with èT and èH as independent variables and 30 effective parameters gave the most accurate prediction on independent emergence data (RMSE=0.078). Traditional non-linear models showed a lower predictive capability (RMSE=0.132– 0.204) irrespective of the index used. These results indicate a higher accuracy and generalization performance of the optimal ANN model in comparison to traditional non-linear regression further suggesting the potential applicability of the former in weed management decision support systems.èT and èH as independent variables and 30 effective parameters gave the most accurate prediction on independent emergence data (RMSE=0.078). Traditional non-linear models showed a lower predictive capability (RMSE=0.132– 0.204) irrespective of the index used. These results indicate a higher accuracy and generalization performance of the optimal ANN model in comparison to traditional non-linear regression further suggesting the potential applicability of the former in weed management decision support systems.