كليدواژه :
رگرسيون , شبكه عصبي مصنوعي , ميكرولايسيمتر وزني , مراحل رشد
چكيده لاتين :
Introduction: In two last decades, greenhouse cultivation of different plants has developed among Iranian farmers, approximately 45 percent of national greenhouse cultures consisting of cucumber, tomato and pepper. As huge amounts of agricultural water in Iran are extracted from groundwater resources and a large number of Iranian plains are in critical conditions, and because irrigation is the major consumer of water (95 percent), it must be performed in a scientific manner. One approach to this is to obtain the knowledge of the consumptive use of major crops which is named evapotranspiration (ETc).
Materials and Methods: This research was carried out in a north-south greenhouse belonging to Plant Protection Research Institute, located on northern Tehran, Iran, for estimating greenhouse cucumber evapotranspiration. Trickle irrigation method was used, and meteorological data such as temperature, humidity and solar radiation were measured daily. Physical and chemical measurements were conducted and electric conductivity (EC) and pH values of 3.42 dsm-1 and 7.19, respectively, were recorded. Soil texture and bulk density were measured as to be sandy loam and 1.4 gr cm-3, respectively. In order to measure the actual evapotranspiration, cucumber seeds were also cultured in six similar microlysimeters and irrigation of each microlysimeter was based on FC moisture. If any drained water was available, it was measured. Finally, with measured meteorological characteristics in greenhouse which are suggested to have an effect on ET and were measurable, the best multiple linear regression and artificial neural network were established. The average data from three microlysimeters were used for calibration and that from three other microlysimeters were used for validation set.
Results and Discussion: In the former case, when we used one multiple linear regression with measurable meteorological variables inside the greenhouse to predict cucumber ET for the entire growth period, high and considerable amounts of error occurred, as the difference between measured and predicted values of ET is approximately 2.86 mm day-1 which is noticeable. Overestimation of the cucumber ET in the first and last stages which will result in decreasing water use efficiency and underestimation in blooming and yielding fruit stages, when cucumber is more susceptible to water stress, are the other disadvantages of using one equation for the entire growth period to describe and predict cucumber ET. In contrast, when we divided growth period into four steps, the MLR method’s performance in prediction of ET was improved and the difference mentioned above between measured and predicted values of ET (2.86 mm day-1) decreased to about 1.32 mm day-1. The results showed that measured and predicted values of ET ranged from (0.08 to 4.75) and (0.13 to 4.25) when the whole growth period is considered as one step, respectively. These mentioned values were obtained (0.08 to 1.5) and (0.13 to 1.75); (0.71 to 2.64) and (1.31 to 4.25); (2.18 to 4.75) and (1.69 to 4.13); (1.32 to 2.61) and (2.66 to 3.74) for each of growth period stages, respectively. Also the value of total ET for the entire growth period is measured 273.45 mm and predicted 275.7 and 275.59 mm, when the whole growth period is considered as one step or divided into four stages, respectively. Although dividing the growth period improved ET prediction, the results in the first and especially the third stage are still discussable. Therefore, as with MLR method, the capability of ANN technique was investigated in prediction of cucumber ET. Comparison of measured and predicted values of ET confirms that ANN has better performance than MLR, even when growth period is divided.
Conclusion: Determining cucumber evapotranspiration in the greenhouse was the main objective of this study. For this purpose we used Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) techniques. In MLR, first we used one equation for the entire growth period. The results showed that this single equation is not able to simulate actual ET of cucumber. To overcome this problem, we divided the growth period into four stages and derived a separate equation for each stage. The results showed that this procedure improves prediction of cucumber ET, especially in the second and last stages of growth period. Statistical indices such as RMSE, Ens, PBIAS and PSR, t-statistical results, measured versus predicted ET values, and predicted values of ET in the growth period indicate that ANN technique is not only reliable, but also easier than the MLR technique.
