Global cross-station assessment of neuro-fuzzy models for estimating daily reference evapotranspiration

• Published in: Journal of hydrology (Amsterdam) 2013-02, Vol.480, p.46-57
• Main Authors: Shiri, Jalal, Nazemi, Amir Hossein, Sadraddini, Ali Ashraf, Landeras, Gorka, Kisi, Ozgur, Fard, Ahmad Fakheri, Marti, Pau
• Format: Article
• Language: English
• Subjects: agricultural management; Artificial neural networks; Assessments; Calibration; climatic zones; Earth sciences; Earth, ocean, space; Estimates; Estimating; Evapotranspiration; Exact sciences and technology; Fuzzy logic; Generalization; Hydrology; Hydrology. Hydrogeology; irrigation scheduling; meteorological data; Neuro-fuzzy model; Reference evapotranspiration; Water resources; weather stations
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2012.12.006

► We used neuro-fuzzy (NF) technique to model daily reference evapotranspiration. ► A global cross station assessment of NF model was performed. ► NF model was generalized (GNF) in humid and non-humid regions. ► Results confirmed the superiority of GNF models to the corresponding equations. Accurate estimation of reference evapotranspiration is important for irrigation scheduling, water resources management and planning and other agricultural water management issues. In the present paper, the capabilities of generalized neuro-fuzzy models were evaluated for estimating reference evapotranspiration using two separate sets of weather data from humid and non-humid regions of Spain and Iran. In this way, the data from some weather stations in the Basque Country and Valencia region (Spain) were used for training the neuro-fuzzy models [in humid and non-humid regions, respectively] and subsequently, the data from these regions were pooled to evaluate the generalization capability of a general neuro-fuzzy model in humid and non-humid regions. The developed models were tested in stations of Iran, located in humid and non-humid regions. The obtained results showed the capabilities of generalized neuro-fuzzy model in estimating reference evapotranspiration in different climatic zones. Global GNF models calibrated using both non-humid and humid data were found to successfully estimate ET0 in both non-humid and humid regions of Iran (the lowest MAE values are about 0.23mm for non-humid Iranian regions and 0.12mm for humid regions). non-humid GNF models calibrated using non-humid data performed much better than the humid GNF models calibrated using humid data in non-humid region while the humid GNF model gave better estimates in humid region.