Improving short‐range high‐resolution model precipitation forecast skill using time‐lagged ensembles

High‐resolution forecasts can vary considerably from run to run. Excessive inconsistency is undesirable, especially for the forecaster, who seeks similarities between successive runs to gain confidence in model guidance. The Met Office 4 km Unified Model runs every 6 h to t + 36 h. Five‐member time‐lag ensembles are created for the most recent 6 h period in order to study the perceived inconsistencies between runs. Mean, maximum, and probability‐of‐precipitation forecast fields are generated, for four different months of operational forecasts chosen to capture different seasonal weather regimes, and compared with the verifying radar accumulation. Each of the individual forecasts (ensemble members) is evaluated and its contribution to the ensemble assessed. Results show that the time‐lag ensemble performs better at or near the grid scale, and that there is a strong case for recalibration. Indeed, the ensemble approach ensures that the proportion of missed forecast events (which have potentially more serious implication for high‐impact weather) is reduced. In short, the effect of the lagged‐ensemble spread is similar to that of the spread created from perturbed initial conditions in conventional ensembles, improving forecast skill at a higher spatial resolution, thus emulating the requirement for spatial averaging for a single deterministic forecast. Overall, improved forecast skill at longer ranges ensures that time‐lagged ensembles are now a more viable option than in the past.