Evaluating Reference Evapotranspiration Using Data Mining Methods and Comparing it with the Results of Water Requirement System in Qazvin Province

The accuracy of determining reference evapotranspiration (ET0) is an important factor in estimating agricultural and garden water requirements. The complexity of the evapotranspiration process and its dependence on meteorological data have made it difficult to accurately estimate this variable. Non-linearity, inherent uncertainty and the need for diverse climatic information in ET0 estimation have been the reasons that have made researchers interested in data mining methods such as artificial neural network (ANNs), random forest (RF) and support vector machine (SVM). Dos et al. (2020) evaluated the performance of machine learning methods to estimate daily ET0 with limited meteorological data. Their results showed that machine learning methods estimate ET0 with high accuracy, even in the absence of some variables. The use of artificial intelligence models in estimating ET0 with high accuracy has become popular in recent years, but the complexity of these models makes it difficult to apply them to regions with different climatic conditions) Feng and Tian, 2021.( Therefore, the aim of this study is to show that different data mining methods are suitable for daily ET0 estimation, which can reach a comprehensive and simple model with high accuracy by using minimal weather data.

In this research, the accuracy of data mining methods in estimating ET0 was evaluated in comparison with the plant water requirement system (FAO-Penman-Monteith standard method). For this purpose, data related to meteorological parameters such as sunshine hour, air temperature, wind speed, and relative humidity air were collected from ten synoptic stations and five climatology stations of Qazvin province in a period of 10 years (1389-1399). The ET0 extracted from the plant water requirement system was calculated based on the Penman-Moanteith method of FAO 56 and on a daily time scale, which is the actual value (measured) with the estimated values obtained by data mining methods (ANNs, RF and SVM) were evaluated. In order to validate the obtained results, the data of each station was divided into two sets of training (two-thirds of data) and testing (one-third of data). Finally, the generalizability of the mentioned methods in estimating ET0 was investigated based on NRMSE, R2, RMSE, MBE, EF and d Criteria.

The results showed that the ET0 values of the plant water requirement system have a good correlation with the estimated ET0 values of ANNs, RF, and SVM methods. In this research, the accuracy of the results of ANNs method was relatively higher than the other two methods. The results of statistical investigations and diagrams showed that ANNs, RF and SVM methods, considering all meteorological parameters (mean air temperature, average relative humidity, sunshine hours and wind speed) as input to the model, in Qazvin synoptic station with altitude 1279 meters and the climatology station of Rajaei power plant with a height of 1318 meters, estimated ET0 with higher accuracy in both training and testing steps.In the ANNs method, the values of NRMSE and R2 at Qazvin synoptic station in both training and testing steps are equal to 0.11 and 0.97, respectively, and at Rajaei Power Plant climatology station in both training and testing steps are equal to 0.10 and 0.97, respectively. In this research, the accuracy of estimating the value of ET0 in two ANNs and RF methods is close to each other and higher than the SVM method. On the other hand, the fitting speed of the ANNs method is very long compared to the RF method, and considering all aspects, it can be said that the RF method has a more suitable approach for estimating the ET0 value. The results of this research showed that the value of ET0 is not only based on air temperature, but may change under the influence of other factors such as air pollution, and is also strongly influenced by regional conditions such as topography and altitude.

The results of this research, in addition to better investigation of ET0, help to know more influential factors in each region and can be used in regions with similar climatic conditions. For example, in the current study area, it was found that the role of average air temperature is greater than other climatic parameters and has a greater impact on ET0. Therefore, it can be said that increasing the average daily air temperature will increase ET0 and subsequently increase the water requirement of plants. As a result, by using these methods and paying attention to these points, it is possible to avoid water stress and possible reduction of the production.