Abstract
The analog method relies on the subdaily distribution of analog days - also called fragments - to perform temporal disaggregation. The features that serve to define analog days are generally at the daily scale although they are assumed to convey information on the subdaily scale. In most cases, the only feature used is the daily total at the location of interest. Features pertaining to the synoptic circulation are also often used although with the purpose of increasing the spatial rather than the temporal resolution. Precipitation is conditioned not only on synoptic circulation but also on regional features such as orography. This combined effect is reflected on the spatial pattern of daily precipitation. We propose to rely on descriptions of the spatial pattern to select analog days in order to perform temporal disaggregation. Three analog method variants based on different sets of spatial pattern features, a variant that rely on synoptic circulation alone and a variant that combines synoptic circulation with spatial pattern features are evaluated and compared. All the features considered characterize a region therefore the same analog day is employed at all the locations within the region. Inter-site coherence is thus enforced. A leave-one-out procedure is carried out on hourly radar precipitation reanalyses covering a mountainous watershed in the French Mediterranean on a 1 km(2) grid over the period 1997-2006. Uncertainty is taken into account by generating five realizations of each variant. Performance is measured in terms of return levels (1-20 years) at various temporal scales (1 h to 12 h) and at two spatial scales - grid box and watershed. The proposed regional approach to temporal disaggregation globally preserves inter-site coherence. This can be seen from the relatively good reproduction of the regional pattern of return levels of the study area by all variants. Synoptic circulation conveys less information on the subdaily scale than spatial pattern features. The variant that depends on synoptic circulation alone largely overestimates return levels especially at lower temporal scales (1 h and 3 h). Combining synoptic circulation with spatial pattern features yielded no significant improvement.