Performance Evaluation of Artificial Neural Networks and Adaptive Neuro-Fuzzy Inference System in Estimating Reference Evapotranspiration under Greenhouse Conditions

One of the important factors for agricultural irrigation management and planning is accurate estimation of reference evapotranspiration (ETo). The application of the standard model (FAO-Penman-Monteith) to estimate ETo is restricted due to the availability of climatic variables including temperature, humidity, radiation, wind speed as well as the availability of the mentioned hypotheses in FAO 56. Accessibility to all climatic parameters or satisfaction of the FAO-Penman-Monteith assumptions are often not possible in some areas or in controlled environments (greenhouses). Therefore, it is preferable to use methods that can provide an accurate estimate of ETo with fewer input parameters. The aim of the present study was to evaluate the performance of two methods of artificial neural networks (ANNs) and adaptive neuro-fuzzy inference system (ANFIS) for estimating ETo in the research greenhouse of the College of Agriculture and Natural Resources of the University of Tehran, located in Karaj, Iran. Based on the measured climatic parameters inside the greenhouse, different combinations were created and evaluation indicators were calculated for each method and scenario. The best neural network structure was obtained for Scenario 4 (radiation, temperature, humidity) with 7 neurons in the hidden layer and Bayesian Regularization training algorithm. ANFIS model was designed with different membership functions. The results showed that there was no significant difference between the performance of the ANFIS method under different scenarios. In other words, even with temperature and humidity data, ETo can be simulated with high accuracy by ANFIS method. Comparison of evaluation indicators between ANFIS and ANNs models showed that ANFIS performed better than ANNs method. The calculated relative root mean square error (RRMSE) for scenarios 1 to 4 in ANNs model was equal to 12.70, 2.23, 2.12 and 2.10% however it was equal to 1.41, 0.80, 1.06 and 1.01% in ANFIS model.