Evaluation and modification of Aphrodite daily precipitation network in Golestan province

Rainfall estimation in regions facing lack meteorological stations is critical in hydrology and meteorology studies. The problem in Iran that lacks dense weather station network, especially in the highlands, is more important. Aphrodite is a dense network that provides daily precipitation data in a resolution of 0.25*0.25 degree. But before operating network, its accuracy and uncertainty should be evaluated. About this database, few studies have been conducted in Iran and the world. This study aimed to evaluate the accuracy of Aphrodite in Golestan province at daily, monthly and yearly scales. Also its accuracy compared with two conventional interpolation techniques and for more data operation, the network was modified.
In this study, 19 rain gauges and 5 synoptic stations were used to evaluate the Aphrodite network. Grid points network were selected based on the nearest point to meteorological station and data analysis was performed in daily, monthly and yearly scales. After accuracy evaluation of network, the correction factor of means ratio was employed for modification of monthly and annual rainfall network data. As well modified network data were compared with interpolation methods of Ordinary Kriging (OK) and Inverse Distance Weighting (IDW). Cross-validation technique was used for evaluation and error indices was calculated for comparison of different methods.
The results of MBE in all studied scales showed that in spite of good correlation of Aphrodite and meteorological data, Aphrodite data has large underestimation error that this is higher in high rainfall areas. Obtained modification factors of the monthly and annual data vary from 1.2 to 1.8 in cold and warm months, respectively. In addition to removal underestimation error from modified network, the average of error is reduced remarkably. Also RMSE of annual rainfall data, is declined from 247 to 194 mm. considering that, network modification needs to be multiplied in a constant (correction factor) so the correlation coefficients remained unchanged. Comparison of modified data network with interpolation methods showed that modified network is more efficient than both interpolation methods. Correlation coefficients of actual and estimated data showed that the correlation of modified network with actual data is positive in all the months but in case of IDW and OK, correlation coefficients was negative in some months that represent modified network in addition to fewer error also in the detection of low and high rainfall points is more efficient than the interpolation methods.
This study showed that high resolution Aphrodite network has good accuracy in rainfall estimation especially in high elevation regions. Also this network is more accurate than interpolation methods of OK and IDW. But because of its underestimation error, before than operating Aphrodite data, the network must be modified.