Estimating transpiration rates of hydroponically-grown paprika via an artificial neural network using aerial and root-zone environments and growth factors in greenhouses

Abstract

Environmental and growth factors are important variables that affect the transpiration rate of crops, but due to their complex nature, it is difficult to systematically use all these factors to estimate transpiration rates. Application of artificial neural networks (ANNs) can be an efficient way of deriving meaningful results from complex nonlinear data. The objectives of this study were to estimate transpiration rates using an ANN, to compare these estimations with the Penman–Monteith (P–M) equation, and to analyze the estimation accuracy according to cultivation period. Paprika (Capsicum annuum L. cv. Scirocco) was cultivated for two cropping periods in a year. Environmental factors were collected every minute and leaf area index (LAI) as a growth factor was measured every 2 weeks. An ANN consisting of an input layer using eight environmental and growth factors, five hidden layers, and an output layer for transpiration rate was constructed. The estimation accuracy in the ANN was higher than the P–M when using aerial environmental factors, but it was further increased by adding root-zone factors. Using daily average data, ANN accuracy was higher for longer cultivation periods and accompanying data. R2 values were 0.88 and 0.73 in the ANN and P–M for one year, whereas they were 0.84–0.93 and 0.79–0.83 for the individual seasons, respectively. The accuracy of the ANN tended to increase when the time step (data-averaging time unit) decreased to 10 min and there was no significant difference over 10 min. Using 10-min average data, the ANN showed high accuracies with R2 = 0.95–0.96 and root mean square error = 0.07–0.10 g m−2 min−1, regardless of cultivation period and season. Therefore, it was confirmed that the ANN could accurately estimate transpiration rates at specific times using the data collected from the entire cultivation period. This approach may be useful for developing irrigation strategies by estimating the transpiration rates of crops grown in soilless cultures.