Foliar photosynthesis estimation in tomato under greenhouse conditions through artificial neural networks

Abstract

Within the theory of systems identification, the models of artificial neural networks (ANN) have shown great ability to find patterns of relationships between variables of highly nonlinear processes, as well as solving the constraint of the nonlinear regression, where it is not possible to use correlated variables as inputs. The objective of this paper was to model the rate of foliar photosynthesis of tomato plants, grown under greenhouse conditions using artificial neural networks, using as input variables: temperature, relative humidity, vapor pressure deficit and concentration of carbon dioxide (CO2) of the air, and photosynthetically active radiation. The experiment was conducted during 2009 in an experimental greenhouse of the Autonomous University of Querétaro, Mexico. The equipment used was the PTM-48M phyto-monitoring (Daletown Company, Ltd), to record CO2 exchange on the leaves, and weather variables. In order to remove the sensor noise in the measurements, the Savitzky-Golay filter was used. Different configurations for back-propagation networks were evaluated, with 4 layers and 10 neurons in the first hidden layer, 15 in the second one and 10 more in the third one, which produced the best statistical indices on the test data: coefficient of determination, R2= 0.9756 and mean square error, MSE= 0.8532. Taking the best ANN predictions, we performed a static optimization, linking two climatic variables with the rate of photosynthesis, using three-dimensional graphics, to show strategies for maximizing the rate of photosynthesis.