We will report statistical findings from a radar climatology of warm season precipitation "episodes" over much of continental North America. Episodes are defined as time/space clusters of heavy precipitation that often result from sequences of organized convection such as squall lines, mesoscale convective systems and mesoscale convective complexes. Episodes exhibit coherent rainfall patterns, characteristic of zonally-propagating structures, under a broad range of atmospheric conditions. Such rainfall patterns are especially prominent under "weakly forced" conditions in mid-summer and these are often concurrent with a monsoon condition over the Rocky Mountain cordillera.

Episodes are initiated in response to diurnal, semi-diurnal, and other forcings. Many events travel eastward across vast stretches of North America (up to 2600 km). A most common longitude of origin is at or near the east slope of the continental divide (105 W). The longevity of episodes (up to 60 h) suggests an intrinsic predictability of warm season rainfall that significantly exceeds the lifetime of individual convective systems and greatly exceeds current predictive skill. Episodes of order 1000 km zonal (E-W) dimension and one day duration are commonplace, since the mean frequency of occurrence from May through August is almost daily. This observation serves both to supplement and diminish the significance of transient forcings and phase relationships at the synoptic scale. It recognizes a continental condition of widespread thermal and hydrodynamic instability and the existence of other processes that routinely excite, maintain and regenerate organized convection on a continental scale.

The propagation speed of major episodes is sometimes in excess of rates attributable either to the phase speeds of large scale forcing or to advection from "steering level" winds. We speculate that wave-like mechanisms, in the free troposphere and/or the planetary boundary layer, may be responsible for such propagation. Such mechanisms may underlie the causes for the observed rainfall coherence and offer the opportunity for markedly improved predictions once these are understood.

A speculative conclusion of this study is that probabilistic precipitation forecasts of 6 to 48 h range can be substantially improved through the joint use of dynamical and statistical methods combined with radar observations. Antecedent convection and its observed propagation routinely place narrow bounds on the future longitudinal position of heavy precipitation episodes, up to 48 h range. Dynamical forecast models routinely identify the latitudinal bands of mesoscale ascent and the associated production of thermodynamic instability. The combined strengths of radar observations, longitudinal prediction by means of statistical expectation, and latitudinal prediction by means of dynamical forecast models may prove to markedly increase skill in warm season heavy rainfall forecasts for the short term prediction ranges. Furthermore, the probability density functions associated with this climatology offer the prospect of improved representation of continental warm season rainfall in global models for applications beyond the deterministic ranges of prediction. This is especially important with respect to the non-local effects of monsoons and topography.