Evaluation of NCEP/NCAR Reanalysis Precipitable Water Data Comparing to Radiosonde Observations for Turkey

Abstract

Precipitable Water (PW) data of NCEP/NCAR Reanalysis Project (NNRP) model is evaluated by comparing to radiosonde data obtained from 8 locations of Turkey for the years between 2015 and 2017. Two methods are utilized to extract NNRP data for the observation locations. In the first method, the nearest NNRP grid point to the radiosonde locations is selected. The second method is the application of bilinear interpolation method on NNRP data to include the weighted effects of corresponding grid locations related with the observation sites. Both NNRP and radiosonde data have 12 h interval for the times 0000 Z and 1200 Z. PW output of NNRP model is compared to observations by means of graphical evaluation of time series, error analyses (Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Root Mean Squared Error (nRMSE)), goodness of fit tests (Cp and PBIAS), and probability density functions (PDF). Error analyses of most of the observation locations indicate that bilinear interpolation method is better than utilizing the nearest grid value data which is not obtained by applying any interpolation technique. Error analyses indicate that nRMSEs of NNRP data for PW analyses are less than 10% for 6 locations of Turkey (Ankara, Diyarbakir, Erzurum, Isparta, Istanbul, and Izmir) if it is assumed that the observations have no errors for the years between 2015 and 2017. nRMSEs of the other 2 coastal locations (Adana and Samsun) are the same as 13.8% and this may indicate that local moisture sources of these locations are greater than mesoscale moisture fields, since NNRP data may not capture local effects well due to its spatial resolution. Comparisons of probability density functions (PDF) of these data sets show that NNRP model may not be successful in capturing extreme values.