عنوان مقاله [English]
نویسنده [English]چکیده [English]
Missing or unsampled data dependent on the hydrological processes are always known as a common and repetitive problem for hydrologists and modelers. Among hydrological variables, rainfall data are critical for hydrological analysis as well as water resources planning and management systems; their estimation is needed in many hydrological modeling studies. In this paper, the potential of artificial neural networks in spatial interpolation of rainfall has been assessed, and also, conventional methods of spatial interpolation are used to have a reasonable basis for the comparison. As a case study, annual rainfall data from Karkheh River Basin in a period of 41 years were used. In order to evaluate the performance of used methods, a comprehensive validation approach is presented based on correlation of each method estimates with the annual discharge of station, which is located downstream of the basin. The results showed that cross-validation approach cannot provide a complete evaluation of the performance of different methods, and the proposed comprehensive approach, provides a basis for determination the suitable spatial interpolation method. According to the results, correlation coefficient of estimates of ordinary kriging (OK), generalized regression neural network (GRNN), inverse distance weighting (IDW) and radial basis function neural network (RBFNN) methods with annual discharge of the basin were equal to 0.846, 0.840, 0.837, and 0.824, respectively, that observed little difference proved the applicability of GRNN and RBFNN methods for spatial interpolation. Generally, it can be concluded that performance of GRNN is comparable to OK; moreover, it needs less predefined parameters and has more clarity and simplicity than OK, give they GRNN more advantage and attractiveness to use.