| Popis: |
Leaf area (LA) is a crucial biophysical variable that is indispensable for many physiological and agronomic models. A reliable and accurate model based on artificial neural networks (ANNs) is proposed to estimate LA of tomato growth under greenhouse conditions. The multi-layer perceptron (MLP) ANN topology was selected for the present study with 5 (ANN5) and three (ANN3) input variables, the architectures were 5-10-1 and 3-9-1, respectively. These MLPs were trained and tested to simulate the response of leaf area with linear measurements leaf length and width. In order to prove the selected topology the ANN was tested with data (leaf length and width) from different experimental growth conditions. Both models had good precision with root mean square errors (RMSEs) of 14.86 and 22.56 cm2, and mean absolute errors (MAEs) of 10.29% and 16.74%, and coefficients of determination (R2) of 0.94 and 0.89, respectively, indicating that the ANN5 model can accurately describe the complex relationship between climate factors (CO2, temperature, and radiation) in different treatments. LA development varied with the different treatments. For high levels of CO2 and temperature, the LA tends to increase highly with respect to the observed area. Variable impact analysis was performed on the input variables; width and length were the variables which impacted the most on LA estimation. Temperature and salicylic acid (SA) concentration were the variables which affect the tomato LA development during the simulations. Overall, ANN models are a useful tool in investigating and understanding the relationships between LA development and climate factors under greenhouse conditions. |
Full Text from ScienceDirect