Climate regionalization with the LMDZ model: a methodological study.

Summary This thesis manuscript reports some methodological studies on climate regionalization by the dynamic approach. The geographical domain covers a large area from the mid-North Atlantic to Eastern Europe, and from the Sahel to the Arctic. The quest for regional climate improvement at all costs is not the focus of the manuscript. However, the focus is on three key points, generally encountered by all attempts to regionalize climate. The first point concerns the nesting scheme: one-way nesting from the GCM to the RCM (OWN) or two-way nesting between the GCM and the RCM (TWN). The second point examines the technical realization of nesting, which is usually a Newtonian relaxation operation added to the prognostic equations of the model. The third point is on the effect of mesh refinement in the RCM. The general spirit of the manuscript is to conceptualize and perform numerical simulations to treat these three points with tricks to isolate and quantify them. The general circulation model LMDZ is used for all the experiments. It plays both the role of the GCM and the RCM. In both cases, it strictly keeps its physical parameterizations and its dynamic configuration, as well as all external forcings or parameters. The experimentation strategy, qualified as Master versus Slave, consists in carrying out simulations under two linked protocols: "DS-300-to-300" refers to downscaling from the GCM at 300 km horizontal resolution to the RCM which is identical to the GCM, also at 300 km spatial resolution. "DS-300-to-100" refers to downscaling from 300 km (GCM) to 100 km (RCM). It is clear that "DS-300-to-300" is an idealized framework, particularly appropriate for evaluating the effect of the relaxation operation. The "DS-300-to-100" protocol, subtracted from "DS-300-to-300", allows for a very accurate evaluation of the effect of the increased RCM resolution. In each protocol, two communication schemes between the RCM and the GCM have been implemented, one (OWN) is the classical one-way methodology of driving the RCM by the GCM outputs, the other (TWN) is to establish a mutual exchange between the two models. Regional climate is sensitive to the choice of communication schemes between the RCM and GCM, especially at mid-latitudes. TWN brings a clear improvement on the representation of boundary information. At the level of regional atmospheric circulation modes, expressed in EOF structures, OWN and TWN are both able to reproduce them, but with slight distortions in space. Newtonian relaxation, widely used in climate regionalization, allows the RCM to track the GCM synoptic path well. However, the temporal concomitance and spatial similarity are dependent on the variables considered, the seasons, the weather regimes, and the spatio-temporal scales of atmospheric circulations. Cases of de-correlation are remarkable when the dominant circulation of the region is of small scales. Mesh refinement increases the freedom of the RCM to develop its internal dynamics, especially at small scales, but also at the whole spectrum of the circulation through the interaction of scales. Thus the RCM becomes more independent and deviates more from the GCM. This thesis, around the methodological aspects of climate regionalization, helps to have a better understanding on the practice. It also sends a cautionary message to the RCM community and invites them to check their regionalization methodology carefully.