In the Adjoint Sensitivity-based Data Assimilation (ASDA) method, an adjoint sensitivity of forecast error is calculated to improve original first guess by generating a perturbation. Dry total energy of forecast error over the specified areas is a response function for the adjoint-sensitivity calculation. The adjoint sensitivity is scaled using a scaling factor which is determined from the minimization of distances between the sensitivity and available observations. The scaled sensitivity is added to original first guess to obtain an improved first guess, and then analysis is made using the improved first guess and observations. This process can give benefit in computational cost. The ASDA method is applied for assimilating radar radial velocity data in the simulation of 10 heavy rainfall cases over the Korean Peninsula. The result is compared with those from 3D-Var and 4D-Var methods.

We evaluated the analysis and subsequent forecast of the ASDA method in terms of radial velocity and precipitation. The Root Mean Square Errors (RMSEs) of radial velocity (the assimilated variable) of the analysis are reduced compared to the first guess in all data assimilation experiments; the RMSE of forecast starting from the ASDA analysis is comparable to that from the 4D-Var analysis, and smaller than that from the 3D-Var analysis. The threat (bias) score of forecast starting with the ASDA analysis is similar to that from the 4D-Var analysis, and greater (closer to one) than that from the 3D-Var analysis. It is also demonstrated that the computational cost of the ASDA method is much lower than that of the 4D-Var method. Limitations of the ASDA method (e.g., more efficient use of the observations with an assimilation window) will be improved in the future work.