Improving Stochastic Rainfall Generators

Stochastic rainfall generators are widely used in hydrologic analysis because they can provide precipitation input to models whenever data are not available. Rather than attempting to reproduce actual rainfall records, stochastic models aim at generating synthetic precipitation time series whose statistics match those of observed ones and, consequently, model calibration consists of determining the generator parameters that minimize the discrepancy between the two sets of statistics. It follows, therefore, that calibration can only be conducted at locations where rain gauges are available and rainfall statistics can be calculated. This paper addresses the following issues regarding stochastic rainfall generators: (1) Seasonality: The generator parameters should vary within the hydrologic year to reflect wet and dry seasons. A total of 12 models (one per month of the year) were developed, each of which had its own parameter set; and (2) Spatial variability. The generator parameters vary from point to point and show spatial patterns in their variability. Maps of the parameters for the entire United States were developed so that synthetic rainfall can be generated at any point in the United States. The application presented in this paper considers, specifically, the Modified Bartlett-Lewis Rectangular Pulse (MBLRP) stochastic rainfall simulation model.