A statistical downscaling to identify the large-scale circulation patterns associated with heavy precipitation events over southern France

This study assesses the potential for a detection algorithm to identify discriminating analysis-based statistical predictors of a few relevant parameters that can be used to capture heavy precipitation events (HPEs), or, at least, their associated large-scale circulation (LSC) patterns in a climate scenario. HPEs are defined from a sample combining 'large-scale' fields from the ECMWF ERA-40 reanalysis with local observations from the Meteo-France rain-gauge network. In a first step, LSC patterns considered as significantly favouring HPE over southern France are identified and described with the greatest robustness possible. For that purpose, an objective automatic clustering of the unfiltered 500 hPa geopotential height field is performed. Four clusters are obtained. Among them, the most discriminating for heavy precipitation is characterised by a synoptic-scale deep upper-level low northwest of the area of interest, inducing a southerly flow over the western Mediterranean Sea and southern France. In a second step, other lower-scale parameters are used to refine the characteristics of the clusters. It has been found that the low-level moisture transport is a relevant low-level ingredient to regionally characterise heavy precipitation. Indeed, 'Cevennes' cases are related to more south to southeasterly flows over the Gulf of Lion, whereas 'Languedoc-Roussillon' events occurred preferentially within a more pronounced easterly wind component with two streams of low-level moisture transport. Moreover, in-depth examination of the low-level features reveals that HPEs tend to occur when the wind blows in a specific direction and for the greatest low-level moisture flux over the Gulf of Lion. Finally, the predictive skill of a detection tool for HPEs over southern France, with only synoptic-scale favourable parameters as predictors, is discussed. It is shown that this tool allows selection of HPE situations in more than 70% of cases. Copyright (C) 2011 Royal Meteorological Society