Extreme Learning Machines: A new approach for prediction of reference evapotranspiration Shafika Sultan Abdullah a,e,⇑ , M.A. Malek a , Namiq Sultan Abdullah b , Ozgur Kisi c , Keem Siah Yap d a Department of Civil Engineering, Universiti Tenaga Nasional, Malaysia b Department of Electrical and Computer Engineering, University of Duhok, Iraq c Civil Engineering Department, Architecture and Engineering Faculty, Canik Basari University, Samsun, Turkey d Department of Electronics and Communication Engineering, and College of Graduate Studies, Universiti Tenaga Nasional, Malaysia e Akre Technical Institute, Dohuk Polytechnic University, Dohuk, Iraq article info Article history: Received 4 February 2015 Received in revised form 27 April 2015 Accepted 29 April 2015 Available online 7 May 2015 This manuscript was handled by Geoff Syme, Editor-in-Chief, with the assistance of John W. Nicklow, Associate Editor Keywords: Reference evapotranspiration Penman–Monteith equation Extreme Learning Machines Artificial Neural Networks summary Recognizing the importance of precise determination of reference evapotranspiration (ET 0 ) is a principal step in the attempts to reserve huge quantities of squandered water. This paper investigates the effi- ciency of Extreme Learning Machines (ELM) algorithm at predicting Penman–Monteith (P–M) ET 0 for Mosul, Baghdad, and Basrah meteorological stations, located at the north, mid, and southern part of Iraq. Data of weather parameters containing maximum air temperature (T max ), minimum air temperature (T min ), sunshine hours (R n ), relative humidity (R h ), and wind speed (U 2 ) for the period (2000–2013) are used as inputs to the ELM model by using four different input cases including complete and incomplete sets of meteorological data. The performance of ELM model is compared with the empirical P–M equation and with feedforward backpropagation (FFBP) model. The evaluation criteria used for comparison are the root of mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ). The statistical results of both models are found to be encouraging; particularly results of running the ELM model with incomplete sets of data, noticing that the sensitivity of the proposed model to missing data changes from one location to another, as well as along the year for certain study location. The R n is found to be the most effective parameter in Mosul Station, while U 2 and R h are found to act almost in parallel with R n in Baghdad Station, and for conditions of Basrah Station; U 2 and R h prove to be the dominant parameters. The minimum and maximum time intervals required for running ELM model for all stations, and in all applied conditions, are (4.64–6.19) seconds respectively, while the same order of timing required for running the FFBP model is (6.30–27.80) seconds. The maximum R 2 recorded for the ELM model is 0.991, while for the FFBP it is 0.985. The ELM proved to be efficient, simple in application, of high speed, and has very good generalization performance; therefore, this algorithm is highly recommended for locations similar to the geographical and meteorological conditions of Iraq that consists of both arid and semiarid regions. Ó 2015 Elsevier B.V. All rights reserved. 1. Introduction Water policy is one of the principal global concerns that aim to control water usage and determine proper ways to minimize the misuse and squandering of available water resources. Evapotranspiration (ET) is a significant variable in the hydrological cycle, and plays a key role in designing and operating irrigation projects; efforts to predict precise ET have been performed by many researchers for nearly a century now, and that is imputed to the early awareness of the importance of water as an essential substance for sustainability of life on this planet. The challenge is further escalated by evolution in urbanization and accelerating growth of population. The need to study and explain the phenomena of evapotranspi- ration stimulated the need for another comprehensive expression called the reference evapotranspiration (ET 0 ). ET 0 is defined as ‘‘the rate of evapotranspiration from a hypothetical reference crop with an assumed height of 0.12 m, a fixed crop surface resistance of 70 sm 1 , and an albedo of 0.23, closely resembling the evapotran- spiration from an extensive surface of green grass cover of uniform height, actively growing, and completely shading the ground with http://dx.doi.org/10.1016/j.jhydrol.2015.04.073 0022-1694/Ó 2015 Elsevier B.V. All rights reserved. ⇑ Corresponding author at: Department of Civil Engineering, Universiti Tenaga Nasional, IKRAM-UNITEN Road, 43000 Kajang, Selangor, Malaysia. Tel.: +60 149341736. E-mail address: sha_akre@yahoo.com (S.S. Abdullah). Journal of Hydrology 527 (2015) 184–195 Contents lists available at ScienceDirect Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol