Radar Rainfall Nowcasting Models Using Moving Motion Vectors

Weather radar plays an important role in enabling short-term forecasting of quantitative rainfall. The most nowcasts of a radar-based surface precipitation pattern is generated by an algorithm based on advection such as the McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation (MAPLE). This method can produce high spatial and temporal resolution results and uses a simpler algorithm than the numerical weather prediction (NWP) model. A disadvantage of the Lagrangian persistence based method is that it can not represent the source-sink or nonstationary states of motion vector fields. In this study, we propose some advection diffusion equation based on the nowcasting rainfall models with moving motion vectors. The diffusion term of this equation leads to smoother predicted images of lead time and increases in some skill scores, and the moving motion vectors are updated each time step by solving a system of two-dimensional (2D) Burgers’ equations.

The procedure of this forecasting model has the following two steps. First, an initial motion vector field is approximated by the Variational Echo Tracking (VET) algorithm. Second, the prediction is obtained at each time step by solving a time-dependent advection or advection diffusion. In this step, motion vectors can be updated at each time step by solving Burgers’ equations. Lastly, in order to verify our method, the results of forecasts are compared with the results of MAPLE. 

High-resolution forecasts for all methods are evaluated for lead times of 2.5 min-3h against rain rate observations for 6 events over 250 km x 250 km region of southeastern Korea. In order to observe the effects of diffusion coefficient and moving motion vectors, the proposed models are classified into four types according to the equation types: advection equation (Type 1), advection equation and Burgers’ equation (Type 2), advection diffusion equation (Type 3), and advection diffusion equation and Burgers’ equation (Type 4). The results of Type 1 is very similar to one of MAPLE. Whereas the other models (Type 2-4) clearly have better skill scores and correlations compared to MAPLE up to lead time of 3h on average.

As results, when the advection diffusion equation was used at low threshold value, the prediction results had good performances than those of advection equation only. Also, when the moving vectors of the Burgers’ equation are used, the forecasts results were better than those of advection diffusion equation or MAPLE. Especially, when the cases are classified as stratiform and convective cases, the results of models using moving motion vectors for convective cases had much higher skill scores than MAPLE. The last model Type 4 generally yielded to better performance than MAPLE and the other presented methods in this study.