Résumé
This study aims at investigating for the first time the links of daily rainfall over Tunisia to large-scale atmospheric conditions by establishing a weather regime (WR) classification over a North African-Mediterranean domain mostly encompassing North Africa. Seven WRs are identified in the year-round ERA-Interim (ERA-I) Mean Sea Level Pressure field spanning 1992–2011. This classification then served as reference to determine WRs in a WRF 10-member ensemble and a regional climate simulations ensemble from Euro-Cordex all driven by ERA-I. Capabilities of both ensemble simulations in reproducing the ERA-I WRs main characteristics, and in reproducing the main features of observed rainfall over Tunisia from a high-resolution local observational dataset and two satellite-gauge merged rainfall products are evaluated for each WR. The models best reproduced the ERA-I two wintertime NAO-like WRs (WR1 and WR3). Better performances of the WRF ensemble and larger spread within the Euro-Cordex models are overall obtained in all WRs. Wet and extremely wet conditions are mainly observed north of Tunisia and seem to be primarily driven by WR5 and WR3 (partly by WR7) featuring strong negative MSLP anomalies and anomalous cyclonic circulation impacting Tunisia. Opposite atmospheric conditions associated with anomalously dry conditions however prevailing all over the country are particularly observed during WR1. Intensity and sign of the simulated rainfall errors over Tunisia, as well as the model uncertainties, as qualified by the Signal-to-Noise ratio, are regime-dependent. The highest (lowest) annual and DJF model errors among the two ensembles are frequently obtained during WR1 (WR3 and WR7) and are discussed with regard to the weakening (strengthening) of the lateral forcing’s magnitude over Tunisia. Model uncertainties are lower in DJF than in JJA. Lateral and orographic forcings restricting inter-member or inter-model spread are mostly associated with WR3 and WR5 in winter. Internal variability of the simulated rainfall field within the WRF ensemble is however persistently smaller than the inter-model spread for all WRs and for all considered timescales.