Application of data assimilation using WRF model to simulate precipitations caused by synoptic systems in the western regions of Iran

Ever increasing attention is paid to numerical weather prediction (NWP) models with the purpose of providing high-resolution precipitation forecasts. In such applications, which are based on both the theoretical analysis and numerical experiments, the prediction accuracy is closely related to the errors in the initial conditions and in the physical parametrization schemes. In the present research, the potential of data assimilation in improving precipitation forecasts was investigated in a case study on an active weather system in the western regions of Iran. Various data assimilation experiments were designed by running the weather research and forecasting (WRF) model and its data assimilation package (WRF-DA). In each data assimilation experiment, we applied the three-dimensional variational data assimilation (3DVAR) method. A heavy rainfall event caused by a strong synoptic system in western Iran was selected in order to study the influence of data assimilation on precipitation forecast.
So as to carry out this study, the initial atmospheric and lateral boundary conditions were taken from three data categories: NCEP global forecast system (GFS), real-time forecasts at 3-h intervals, which are gridded to horizontal resolutions of 1̊×1̊ and 0.5̊×0.5̊, NCEP FNL (Final) Operational Global Analysis data on 1̊×1̊ grids prepared operationally every six hours and ERA-Interim reanalysis dataset of ECMWF, gridded to horizontal resolution of approximately 80 km at 6-h intervals.
Simulations were divided into control runs and data assimilation runs, with the former runs being based on three sets of data as initial conditions. The data assimilation runs were conducted utilizing GFS data as the background and two sets of obeservations, namely the surface observations of Iran Meteorological Organization (IRIMO) and the NCEP observations. The observation data showed a significant impact on the initial conditions of 2m temperature and 10m zonal and meridional wind components, such that in certain parts of the simulation domain, the background temperature was estimated to be up to °relative to the analysis and the wind field was revised by up to ±3 meters per second in some areas.
The comparison between the scatter plots of the background and observations relative to the analysis corroborates the fact that the scatter and errors were decreased after using 3DVAR. The findings indicated that the accuracy of forecasts depends directly on the type of data employed as initial conditions for WRF model and the physical parametrization schemes, hence the fact that the simulations demonstrate significant differences. The bias analysis of precipitation for stations with precipitation records in the west illustrated that the assimilation of IRIMO surface data in one of the physical configurations decreased the forecast bias to a minimum of 73% of the cumulative 24-hour precipitation forecast. The impact of data assimilation, on the other hand, decreased in the cumulative 48-hour precipitation forecasts.
Correlation analysis of the forecasted precipitation patterns and the observed values demonstrated that data assimilation generates a higher correlation coeffcient, implying that it had a discernible, though limited, positive impact on the case examined. In addition, the maximum impact of data assimilation on the correlation between data assimilation runs and control runs for precipitation was approximately 8%. Specifying a precipitation threshold for quantitative precipitation forecasts (QPF), the binary analysis was done, while the proportion correct score (PC) of each threshold was employed in order to investigate the forecasts quality. In conclusion, using the skill score of binary analysis is not a proper method to compare forecasts quality in different experimental runs when the number of forecasts and observational stations are limited.