Reconstruction of Missing Data of Monthly Total Sunshine Hours Using Artificial Neural Networks

Reference crop evapotranspiration is one of the important factors of hydrological cycle. This parameter is used to design irrigation systems¡ hydraulic structures and drainage systems. One of data that required to calculate the amount of reference crop evapotranspiration is solar radiation which in the absence of this data¡ monthly sunshine duration data will be used. At the most of the weather stations of Iran the data of monthly total sunshine hours is not available at previous years¡ so the need to rebuild the data is felt. In the present study two kind of artificial neural network model (MLP and RBF) and meteorological data of target station and monthly total sunshine hours of neighbor stations are used to rebuild the missing data. The results showed that using data from meteorological data of target station and neighbor station¡ the total monthly sunshine reconstructed with high precision. The results of the different scenarios showed that if only the meteorological of target station such as minimum and maximum temperature¡ average relative humidity¡ solar radiation¡ extraterrestrial radiation and straight¡ dark and cloudy days number is used¡ with a precision of RMSE=16.79 hour and MAR=6.44% the monthly total sunshine hours is estimated. Also if only the data from nearby stations is used¡ the estimates would be more conducive to accuracy (RMSE=14.25 hour and MARE=5.71%). The best results were obtained when both weather data set of target station and adjacent stations are used (RMSE=13.78 hour and MARE=4.97%). Comparison of the performance of the ANN-MLP and RBF ANN-MLP showed that the accuracy of MLP neural network is somewhat greater. Finally the time series of monthly total sunshine hours and reference evapotranspiration were renovated.