GENERALIZATIONS OF CHAIN-DEPENDENT PROCESSES - APPLICATION TO HOURLY PRECIPITATION

Stochastic models are fitted to time series of hourly precipitation am ounts. These models are extensions of a form of chain-dependent proces s commonly fit to daily precipitation amounts. The extensions involve allowing hourly intensities to be autocorrelated and allowing the mode l parameters to possess diurnal cycles. These models are applied to tw o quite different sets of hourly precipitation data: July at Denver, C olorado, for which diurnal cycles are substantial; and January at Chic o, California, for which a relatively high degree of persistence is pr esent. The temporal aggregation properties of the hourly models (e.g., for 12-hour or daily total precipitation) are examined, and the role of the extensions in improving these properties is quantified. On this basis, it is argued that generalizations of chain-dependent processes could be competitive with, if not superior to, so-called conceptual m odels of the precipitation process.