Kilometre‐scale numerical weather prediction addresses the challenge of forecasting accurately clouds and precipitation. Ensemble‐based data assimilation methods make use of background‐error covariances that are sampled from an ensemble of forecasts. These methods can be considered in order to include hydrometeor variables and their flow‐dependent error covariances in the data assimilation system. Yet, because of limited ensemble size, rank deficiency of the resulting covariances and sampling noise occur, which can be mitigated by a localization procedure. In order to optimally localize covariances for hydrometeor variables, previous work by the authors has been extended. This approach estimates localization as a linear filtering on covariances, optimal in the sense of minimizing sampling noise. The zero‐variance and the high spatial variability issues met with hydrometeor variables are addressed by using an improved method for spatial sampling, based on geographical masks. Diagnosed optimal horizontal localization lengths appear to be much shorter for hydrometeors than for other classical thermodynamic variables. Conversely, we report optimal vertical localization to be very broad for precipitating species. Great variability between different meteorological situations has also been noticed, which reflects the high flow dependency of hydrometeor forecast errors. This suggests that ensemble‐based data assimilation schemes that consider hydrometeors as control variables should adopt more refined localization schemes than the common “one‐size‐fits‐all” approach.

Traditional pointwise verification scores are not always appropriate for the evaluation of high‐resolution precipitation forecasts because of double‐penalty problems. An alternative approach, based on the identification of homogeneous rainfall areas called “precipitating objects”, allows forecast evaluation at a larger and thus more predictable scale, and specific information about the nature of errors (e.g. location, size, intensity) can be obtained. A novel object detection method is first introduced and the object‐based verification of precipitation forecasts from the convective‐scale deterministic and ensemble models Arome and Arome‐EPS is then discussed, using several scores and diagnostics. Three types of precipitating objects characterizing total, moderate and heavy rainfall are considered. In the second part, object‐based metrics are used to compute objective weights for time‐lagged ensemble forecasts, based on their performance at early forecast ranges. The weights obtained clearly depend on the meteorological situation and on the precipitation type, reflecting for instance the lower predictability of moderate precipitation compared to total precipitation. There is also a dependence on the production time with, on average, slightly larger and more homogeneous weights associated with the most recent run. However, in some situations of moderate and heavy rainfall, a relevant signal can be extracted from older runs. It is finally shown that object‐based weights are better suited than classical quadratic weights to improve nowcasting performance.

Abstract. A new object-oriented method has been developed to detect hazardous phenomena predicted by Numerical Weather Prediction (NWP) models. This method, called similarity-based method, is looking for specific meteorological objects in the forecasts, which are defined by a reference histogram representing the meteorological phenomena to be detected. The similarity-based method enables to cope with small scale unpredictable details of mesoscale structures in meteorological models and to quantify the uncertainties on the location of the predicted phenomena. Applied to ensemble forecasts, the similarity-based method can be viewed as a particular case of neighborhood processing, allowing spatialized probabilities to be computed. An application to rainfall detection using forecasts from the AROME deterministic and ensemble models is presented.