Abstract
The impact of variational data assimilation for predicting two heavy rainfall events that caused devastating floods in Jeddah, Saudi Arabia is studied using the Weather Research and Forecasting (WRF) model. On 25 November 2009 and 26 January 2011, the city was deluged with more than double the annual rainfall amount, caused by convective storms. We used a high-resolution, two-way nested domain WRF model to simulate the two rainfall episodes. Simulations include control runs initialized with National Centers for Environmental Prediction (NCEP) Global Forecasting System (GFS) data and three-dimensional variational (3D-Var) data assimilation experiments conducted by assimilating NCEP PREPBUFR and radiance observations. Observations from Automatic Weather Stations (AWSs), synoptic charts, radar reflectivity and satellite pictures from the Presidency of Meteorology and Environment (PME), Jeddah are used to assess the forecasting results. To evaluate the impact of the different assimilated observational datasets on the simulation of the major flooding event of 2009, we conducted 3D-Var experiments assimilating individual sources and a combination of all datasets. Results suggest that while the control run had a tendency to predict the storm earlier than observed, the assimilation of profile observations greatly improved the model's thermodynamic structure and led to better representation of simulated rainfall both in timing and amount. The experiment with assimilation of all available observations compared best with observed rainfall in terms of timing of the storm and rainfall distribution, demonstrating the importance of assimilating different types of observations. Retrospective experiments with and without data assimilation, for three different model lead times (48, 72 and 96 h), were performed to examine the skill of the WRF model to predict the heavy rainfall events. Quantitative rainfall analysis of these simulations suggests that 48 h lead time runs with assimilation of all observational data provide the best statistical scores.