Characterization of the error structure of precipitation nowcasting

Radar-based extrapolation techniques based on Lagrangian persistence have shown skill to forecast precipitation for lead times of several hours. However, these methods are affected by two main sources of error: (i) the temporal evolution of the rainfall field (growth and decay), which is the most significant, and (ii) the estimation and evolution of the motion field. Errors in rainfall nowcasts have a direct effect on applications such as hydrological flow forecasting. In quantitative precipitation estimation (QPE), the generation of rainfall ensembles of equiprobable rainfall field scenarios is seen as an interesting approach to account for the errors affecting radar-based rainfall maps and their propagation. Furthermore, the ensemble methodology can be applied to quantify the uncertainty of radar-based nowcasts. For this purpose, a characterization of the nowcasting error structure is needed.

The aim of the present study is to characterize the errors affecting the forecasts produced by the McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation (MAPLE). The description of the structure of the errors has been carried out over several months of MAPLE nowcasts for the Central and East United States and it includes a description of the statistical distribution of the errors and their correlation in space and time. In this characterization we have investigated the dependence of the error on the location (results show that the distribution significantly changes from region to region), the time of the day (to quantify the impact of the diurnal cycle of precipitation) and on season.