Using Unsupervised Estimator Technique to Predict Reference Crop Evapotranspiration

Evapotranspiration is the main component of hydrologic cycle and has an important role in crop water requirement estimations, water balances studies, and water resource management. There are a lot of direct and indirect methods to estimate reference crop evapotranspiration, but each has some limitations. For example, limitations that can be mentioned for direct measuring are the insufficient precision in measuring devices and the scale problems. An indirect method like Penman-Monteith on the other hand needs a lot of daily climatic parameters. This research tried to use self-organizing maps as an unsupervised artificial neural network method to predict evapotranspiration by minimum meteorological data input. Based on fuzzy clustering indices, evapotranspiration values in the study area, Mashhad plain, are divided into two clusters with low and high ETo coincided with the climate of the area. Also, in order to validate the model, statistical indices containing root mean square error, determination coefficient, and Nash–Sutcliffe model efficiency coefficient are used and the results are compared with the experimental models output. The results showed that even the simplest SOM model which employs mean temperature and maximum sunshine duration as input have less errors compared to the experimental equations.